evaluation of fingerprint recognition

Study: “Evaluation of Fingerprint Recognition Technologies – BioFinger“

Public Final Report

Version 1.1 06.08.2004


Content Study: “Evaluation of Fingerprint Recognition Technologies – BioFinger“ 1

1 Summary 5 1.1 Objective 5 1.2 Results 6 1.3 Structure of the Report 8

2 Biometric Authentication with Fingerprint Recognition Systems 9 2.1 Introduction 9 2.2 Requirements on a Biometric System 11 2.3 Operative Capability of a Biometric System 12 2.4 Fingerprint Recognition 13 2.4.1 Prob lem Definition 13 2.4.2 Fingerprint Scanning 14 2.4.3 Pattern Classification 16 2.4.4 Fingerprint Image Comparison 18 2.4.5 Image of the Fingerprint Identification Procedure 20 3 Evaluation of Biometric Systems 22 3.1 Description of the Evaluation Criteria 22 3.1.1 Types of Errors 22 3.1.2 Objective Comparison of Fingerprint Systems 25 3.2 Experimental Determination of the ROC curves 26 3.2.1 Determination of the Probability Density Functions 26 3.2.2 Calculation of FNMR(T) and FMR(T) 26 3.2.3 Determination of the ROC curves 27 3.3 Police-Related Application Scenarios of Biometrics Systems and their

Requirements regarding Error Rates 28

4 Investigations with Test Persons 30 4.1 Inclusion of the Database 30 4.1.1 Sensors and Algorithms 30 4.1.2 Description of the Sensors 31 4.1.3 Description of the Algorithms 43 4.2 U1 – Influence of the Sensors on Verification 47 4.3 U2 – Influence of Feature Extraction on Verification 48 4.4 U3 – Influence of the Algorithms (MSA) on Verification 50 4.5 U4 – Influence of the Sensors on Fingerprint Image Quality 53 4.5.1 Contrast 53 4.5.2 Average Value of Grayscales 53 4.5.3 Separability 57

2 Federal Office for Information Security


4.6 U5 + U6 – Influences on the Fingerprints 58 4.6.1 U5 – Influence of the Sensors on the Fingerprints 58 4.6.2 Influence of Feature Extraction on the Fingerprints 60 5 Test Results of Various Systems 62 5.1 Introduction 62 5.2 Evaluation of the Fingerprint Quality 63 5.3 Comparison of the Systems 65 5.4 ROC Curve for Sensor 1 69 5.5 ROC Curve for Sensor 2 70 5.6 ROC Curve for Sensor 3 71 5.7 ROC Curve for Sensor 4 72 5.8 ROC Curve for Sensor 5 73 5.9 ROC Curve for Sensor 6 73 5.10 ROC Curve for Sensor 7 74 5.11 ROC Curve for Sensor 8 74 5.12 ROC Curve for Sensor 9 75 5.13 ROC Curve for Sensor 10 75 5.14 ROC Curve for Sensor 11 76 5.15 ROC Curve for Sensor 13 76 5.15.1 Description of the System 76 5.15.2 ROC curve 77 5.16 ROC Curve for Algorithm 1 78 5.17 ROC Curve for Algorithm 2 79 5.18 ROC Curve for Algorithm 3 80 5.19 ROC Curve for Algorithm 4 81 5.20 ROC Curve for Algorithm 5 82 5.21 ROC Curve for Algorithm 6 83 5.22 ROC Curve for Algorithm 7 84

6 Investigations with the Fingerprint Database 85 6.1 Description of the Databases 85 6.2 Research on the Differentiability with Similar Fingerprints 87 6.2.1 Description of the Examination 87 6.2.2 Results 89 6.3 Research on the Ageing Characteristics of Fingerprints 93 6.3.1 Description of the Examination 93 6.3.2 Results 95 6.3.3 Examination of Ageing According to Age Groups 103

Federal Office for Information Security 3


6.4 Simulation of the Reduced Sensor Area 105

7 Standards and Universal Fingerprints 110 7.1 Feasibility and Algorithm Methods 110 7.1.1 Feasibility 110 7.1.2 Algorithm Procedures 111 7.2 Documentation of Standards 111 7.2.1 General Standards 111 7.2.2 Fingerprint-Specific Standards 116 8 Bibliography 120

9 Table of Abbreviations 122



1 Summary

1.1 Objective

As a biometric identification property, fingerprints have had a long tradition and are a synonym for the uniqueness (of man). Up until recently, it was only the resulting fingerprint image that was exclusively used as an identification feature; no further processing was carried out. Human fingerprints were almost solely used for forensic purposes in dactyloscopy. Dactyloscopists examine fingerprints with regard to details that can be used to identify people. Evidence of a fingerprint found at a scene of a crime can thus be allocated to a person as the one who left that trace. Since fingerprints can be classified, they can be categorized into various finger classes by making use of the fact that due to the ridge flow so-called patterns (loops, arches, whorls) are formed and that due to the interruptions of the ridges, anatomic characteristics (minutiae) are shaped. Thanks to the large dactyloscopic information content in individual prints, a dactyloscopic expert can determine, by comparison, whether individuals are identical or not. In the past, it took a lot of time to find one person in a hard copy database (identification) and then to prove that the fingerprints at the site of the crime and in the database were identical. The initial use of computers for identification purposes was limited by a quick searching of an electronic database. Dactyloscopic experts provided the details necessary for that searching process. Since computer performance capacities have increased, image processing of fingerprints and thus their electronic evaluation became possible. Initially, dactyloscopic systems analyzed and extracted all known details, i.e. patterns and the set of features. As far as their application in an access control system was concerned, the use of these comprehensive details resulting from fingerprints proved to be impractical. Processing time was too long and the amount of extracted details too large. As a result, the amount of data was reduced, i.e. certain patterns were treated as negligible. Additionally, the number of minutiae was reduced. Mostly, for today’s access control systems, minutiae are simply defined as ridge endings or ridge bifurcations. More recent developments are aiming to use not just the minutiae but also the image information of the surroundings of a particular detail by covering it with a filter mask (e.g. by use of Gabor Filter). As part of the project called “BioFinger – Evaluation of Fingerprint Recognition Systems – Fingerprint Technologies”, the characteristics of fingerprint recognition systems are analyzed. The background of this project is the possible integration of fingerprints in German personal documents in order to improve the verification of the holder of the document (i.e. ID Cards, driver’s licenses, passports). Hence, the very aim of the BioFinger Project is the verification, i.e. the examination of the identity claimed by the person (1 on 1 comparison). On the other hand, with regard to envisaged application, identification (one on x-comparison), with which a person is to be identified by comparing him/her with x number of people in a database, does not play any role. Within this context, a number of examinations are carried out in the BioFinger Project, which are to clarify the suitability of some chosen products. The question is this: Using today’s systems or components, are there fingerprint recognition systems that have verification characteristics, or can they be assembled. Due to the special demands on personal documents, i.e. usable lifetime of ten years, the ageing of fingerprints with regard to their characteristic to identify people, is very significant. At the beginning, a market analysis, which includes all fingerprint technologies available on the German market, is carried out. Furthermore, a number of selected systems of foreign companies are included in the investigation. Promising systems are chosen from this survey. For this purpose, software algorithms and sensor hardware that is used are tested. This examination is meant to clearly show possible significant differences in those fingerprint recognition technologies. The associated ROC (Receiver Operating Characteristic) curves are set up in order to assess efficiency and comparability of the chosen fingerprint recognition technologies.



In addition, the algorithms were examined with regard to their capability of differentiating between so-called biometric twins (persons whose fingerprints were classified as being similar by the fingerprint recognition systems). This was under-girded by examining the influence of the ageing process on the algorithm performance. The Federal Office of Criminal Investigation (BKA) provided the specially selected databases.

1.2 Results

Achievable Recognition Performance The examination has shown what kind of performance today’s technology can achieve. The result was that half of the tested systems 1 had an EER2 [Equal Error Rate) of less than 5%. One third achieved EERs below 3%. In the top range (EER ≤ 1%), there are 8% of the tested systems.

As far as the verification of passport or identity card holders is concerned, the recognition system will probably be run in such a way as to have an FAR3 [False Acceptance Rate] that is better than the EER, e.g. FAR = 1‰. Even though this leads to a worse FRR4 [False Reject Rate], half of the tested systems still generates FRRs below 10% for this operational mode. About 23% of the tested systems can still reach FRRs of 3% or less.

1 Combination of scanner and algorithm 2 EER: Equal Error Rate; see 3.1.1 for definition 3 FAR: False Acceptance Rate; see 3.1.1 for definition 4 FRR: False Reject Rate; see 3.1.1 for definition



Number of test pieces

0 2 4 6 8

10 12 14

0% 10%

20%

30%

40%

50%

60%

70%

FFR with FAR=1/1000

abso

lute

0%

20%

40%

60%

80%

100%

cum

ulat

ed

This means that, if mutually compatible components (scanner and algorithm) are carefully chosen, only one out of 1,000 persons with a false ID card would, despite his/her false identity, be accepted by the biometric system. However, the probability of wrongly rejecting a person with a correct ID card would be about 1:50. Thus, this technology shows an effective improvement to people comparing faces with ID card pictures.

Influence of Components

A few combinations of sensors and algorithms led either to a notably higher error rate or were not compatible at all. Comparing results of individual sensors showed significant differences. For instance, the best sensor achieved an error rate that was ten times lower than the worst one. Optical sensors operating with the method of frustrated total internal reflection achieved the best results. Differences between algorithms were notably less pronounced. The best algorithm achieved error rates that were three times lower than those of the worst algorithms. Influence of the Age of Reference Data The possible use of fingerprints in personal documents raises the question of whether recognition ability stays the same if reference and verification data were not recorded within a certain period of time but rather at large intervals. In principle, the wider the time frame, the worse the FRR that is to be expected. Based on the examinations that were carried out, it can be estimated that the FRR doubles if the time period reaches ten years. Standards and Universal Fingerprints

The different templates of the various algorithms for recognizing fingerprints showed a great variety in design despite the fact that they had some features in common. Some systems extract only minutiae; others, however, additionally extract patterns or else they use image information either exclusively or in addition. As a consequence, one system would, under normal circumstances, not generate the details, which are needed by the other system for its verification purposes at its usual level of performance. Although the smallest possible common amount of data with which all algorithms can work guarantees the inter-operability of various individual systems, it would, nevertheless, have negative effects on the biometric recognition performance and increase error rates. Highest inter-operability without a decrease in performance can be reached if fingerprint images are stored instead of features. However, this requires additional feature extraction for every single verification procedure.



Possibilities for Improvement

Since this examination was carried out with all fingers except for the small finger, further improvement can be expected if only fingers with a large area (thumb, index finger) are used. Since not only the best fingerprints were used but rather all images were analyzed, an improved recognition performance can be expected, if for example in case of a wrong rejection further verification attempts are allowed for or if a quality control is carried out at enrollment. The follow-up study, BioFinger2, shall show what kind of improvement can be reached if several fingerprints are used for verification purposes.

1.3 Structure of the Report

The individual chapters of the report are briefly described below. The second chapter describes biometric systems in general. At the beginning, there is a short introduction into the topic of biometrics and a description of a biometric system is given. The requirements for biometric systems are explained and performance parameters of a biometric system are defined. In the second half of the chapter, the fingerprint recognition procedure is discussed in detail. The third chapter elaborates on the evaluation criteria of biometric systems as well as the definition of such criteria. Furthermore, evaluation metrics are linked to concrete police-related application scenarios. Thus, the concrete feasibility of the tested system can be evaluated with regard to its intended purposes. The fourth chapter describes the examination of fingerprints from test persons. It is introduced with the description of the recording of the database and contains a table of sensors, the set-up of the database from the fingerprints of various persons as well as an explanation of the database analysis in order to filter out errors. Within the framework of the study, a number of examinations (E1 through to E6) are carried out, which are also mentioned in this chapter. U 1 Influence of the sensor quality on the verification quality (FAR, FRR)

U 2 Influence of the quality of feature extraction methods (PE) on the verification quality

U 3 Influence of the various matching systems (MSA) on the verification quality

U 4 Influence of the various sensors on the quality of the fingerprint images (resolution, etc.)

U 5 Influence of the sensor quality on the quality of generated feature vectors

U 6 Influence of the various PEs on the generated feature vectors. Which features are extracted by an algorithm (e.g. "Only coordinates of the minutiae" or "coordinates of minutiae and directions", number of features, data quality of features, etc.)?

The fifth chapter summarizes the results of the above-mentioned examinations for various sensors and algorithms. For the purpose of this test, eleven sensors and seven algorithms were used. The sixth chapter contains a description of the investigations carried out using the fingerprint image database provided by the Federal Office of Criminal Investigation (BKA). The seventh chapter describes today's standards (with regard to fingerprint recognition technologies); it also discusses the possibility of a universal fingerprint standard. Darmstadt, 20 May 2004



2 Biometric Authentication with Fingerprint Recognition Systems

2.1 Introduction

In order to classify fingerprint recognition as a biometric procedure, the following terms need to be defined: [BRO02], [TTT02]: • Static features are anatomic characteristics of the body, which either change very little or not at

all in the course of life (fingerprints, eye color, iris, genetic data, etc.). • Dynamic features are behavioral characteristics of man (handwriting, walk, voice, etc.). • Passive acquisition is "pass by" acquisition (e.g. of a face by a camera). • Active acquisition describes an acquisition process involving the person (i.e. fingerprint). • Identification: Establishing of identity (1 on x-comparison; who is this person?). For

identification purposes, the biometric feature is compared with all reference details stored in the biometric system. If any characteristics match, the identification process was successful and the corresponding name (e.g. User ID) belonging to this reference feature can be processed further.

• Verification: Confirmation of identity (1 on 1 comparison; is this person who she/he claims to be?). For verification purposes, the user states his/her identity to the biometric system in advance (e.g. his/her User ID is entered via a keyboard or chip card). Then the system only has to compare the biometric feature with one reference feature that matches the User ID. If they are identical, the verification process was successful. Verification is done significantly faster than identification if the number of reference features / users is very high. At the same time, verification is much more reliable than identification, especially if the number of reference features is very high.

• Authentication: Attestation of genuineness (proof of identity, e.g. by identification or verification).

• Authorization: Authorization means "empowerment.” Following a successful authentication (identification or verification) using a biometric system, a person is given permission to carry out certain actions or to use certain services.

• Biometric system: Biometric recognition systems process biometric features of a person with the aim of confirming or rejecting that person's identity by using previously gathered reference data. In general, all biometric systems are made up of the following components: data input, pre-processing, feature extraction, classification, and calculation of reference data. For adapting to changes in the biometric pattern, an adaptive procedure can be used. Figure 2.1 demonstrates the basic set-up of a biometric system. Figure 2.2 shows the verification process. Data input is carried out via a sensor. The data is pre-processed and normalized prior to and during the comparison of patterns. For classification purposes (i.e. for categorizing fingerprint image types into given finger classes) both pre-processed data or extracted features can be used. The initial input data or features are compared with respective reference data. In order to choose reference data in the reference database, the user may, for example, indicate his personal identification number. As an alternative, reference data may also be stored on a storage medium, such as a chip card, which the user holds. As far as adaptive procedures are concerned, if the classification was positive, the results thus achieved can be used for updating reference data.

Nowadays, the demand for reliable identification procedures is increasing. Currently, we encounter the issue of personal identification e.g. in e-commerce, access control facilities, in the fight against terrorism etc. Even though identification by means of an object, e.g. an identity card, is still fulfilling its purpose, it is continually loosing its importance in our modern, electronically communicating world of more than 6 billion people. For this reason, biometry has, especially in recent times, been getting



more important since it combines personal identification with unambiguous and unchangeable characteristics of man. With ever-increasing and evermore complex technologies, exact personal identification is imperative. By using identification processes, it is for example possible to regulate access to certain objects by granting certain rights. Everyone who was positively identified and thus accepted is given pre-established privileges. In the police, identification (e.g. dactyloscopy) plays an important role. These are only two of many cases where "biometric" identification is used. Man has certain unambiguous features (in the sense of physical characteristics) which are formed in the earliest phases of human life as part of a random process (randotypical) and which are different for each individual. One of the first biometric features that was discovered and scientifically investigated was the fingerprint. The surface of the inguinal skin of man and of most mammals shows patterns and their variety seems to be endless. For example, the ridges of the inguinal skin on the fingers of humans are different. Ridges form various patterns (loops, arches, whorls) which – in connection with interruptions of the ridges (minutiae) – differ from finger to finger. For forensic purposes, fingerprints were used as early as at the end of the 19th century in order to identify people (dactyloscopy) [HEI27]. With the advancement of technology, the issue of safety has become more important. For access controls, analyzing fingerprints biometrically has been playing an increasingly important role.

Registration module

Biometric sensor

Characteristics

IDENTIFICATION MODULE

Template database

Features Biometric sensor Comparison

Figure 2.1: General biometric system



Data acquisition

Pre-processing

Feature extraction

Data

Feature vector

Reference formation

Reference data Reference data adaptation

Classification

Decision original / falsification

Figure 2.2: Sequence of biometric verification

2.2 Requirements on a Biometric System

Each physiological or behavioral feature can be used as a biometric characteristic for personal identification processes as long as they fulfill the following requirements: • Universality: Every person has to have this feature, • Uniqueness: No two or more people with the same feature must exist, • Constancy: The feature does not change significantly in the course of time, • Collectability: The feature must be measurable or collectable. However, biometric features that are currently being used mostly do not fulfill all of the above-mentioned requirements. Hence, they are only partly suitable for a practical application in biometric systems. In addition, further practical aspects have to be taken into consideration: • Performance of the system which allows for quantitative statements with regard to identification

accuracy and speed as well as the required robustness in the face of system-related factors, • Acceptability of the system in its practical application, • Fake resistance of the system, i.e. robustness against direct methods of tricking the system. Hence, for most applications, practical biometric systems have • to perform with and at an acceptable identification accuracy and speed, • to have reasonable requirements with regard to biometric features, • to be non-invasive, • to be accepted by the users, • to be sufficiently robust against misuse.



2.3 Operative Capability of a Biometric System

Although digitisation and the ensuing reduction of biometric data, which is necessary for its electronic processing, result in a very large classification of various features, it, nevertheless, shows a very rough granularity. For this theoretical reason, the system’s answer whether someone is an authorized person or not, will not be an absolute “yes” or “no”; the answer will be expressed with a certain quantitative “index” or “matching score” instead. An identification result may read as follows: the index, according to which the biometric patterns of person A are identical with the stored data of person A, is 0.85. In this case, the index is the result of the matching algorithm and can, for example, show the degree to which the biometric characteristics between the actual data and the stored reference features are identical. Each matching process, will report an other numerical value which is a reflection of the statistical and system-related variations. By choosing a threshold value for these results, the “right“ identifications are separated from the ”false” ones. An analogous distribution can be made if an attacker (person B) is wrongly identified as person A. In this case, a threshold value also separates "false" and "right" results. These two distributions and a common threshold result in four different cases: • (Person A is correctly identified as A (correct identification), • (Person A is rejected as A (false rejection), • (Person B is rejected as A (correct rejection), • (Person B is accepted as A (false acceptance).

Acceptance threshold

0 100 Correspondence value

Freq

uenc

y

FRR FAR

No correspondence Correspondence

(a) and (c) are correct, (b) and (d) are erroneous cases. If the above-mentioned frequency distributions for (b) and (d) are integrated, i.e. by fixing the threshold value in such a way as to define the upper or bottom integration limit, there will only be two results, which, once they are standardized, show the so-called False Acceptance Rate (FAR) and the False Rejection Rate (FRR). FAR is defined as the probability that a person B is accepted as person A. FRR is defined as the probability that person A is rejected. Due to the overlapping of both distributions, compromises have to be made when fixing the threshold values for the system; a lower FRR usually leads to a higher FAR and vice versa (Figure 2.3). Usually, the performance of a biometric system for automated personal identification is defined by the FAR and FRR factors. For example, if FAR is zero, it means that no attacker was accepted.

Figure 2.3: FAR and FRR

There are additional parameters such as verification and identification speed, which are used to determine the performance of the systems. Due to the “one-on-one” comparison method in verification processes, the speed is mainly limited by the time the computer needs to carry out the verification



algorithm. Usually, it is quite easy to meet the speed requirements in these cases. As for identification processes, however, and especially in systems which have millions of entries the number of required comparisons does limit the overall speed of the system.

2.4 Fingerprint Recognition

2.4.1 Prob lem Definition

n the context of the term "identification through fingerprint images", fingerprints are generally accepted as human fingerprint images. Identification can functionally be split up into the following three basic tasks: • (Fingerprint scanning, • (Fingerprint classification, and • (Fingerprint comparison. Fingerprints can be aquired as color prints or via sensors which store the ridges on a medium (glass, paper, sensor surface, etc.). During the classification process, fingerprint images are optionally allocated to a certain category based on the global orientation of the ridges while the location of the minutiae is marked as well. The comparison determines whether two fingerprint images are identical, i.e. whether they belong to the same person (finger). The complete process of a fingerprint image analysis (comparison of patterns) can be divided into six steps (Confer Figure 2.4).

Scan

ning

of t

he fi

nger

prin

t im

age

Imag

e qu

ality

impr

ovem

ent

Imag

e pr

oces

sing

Patte

rn c

lass

ifica

tion

Feat

ure

extra

ctio

n

Ver

ifica

tion

phas

e

Figure 2.4: Process of fingerprint analysis

1. Scanning of a fingerprint image. The quality of the scanned image is the decisive factor for automatic identification purposes. It is desirable to use a high-definition fingerprint scanner which is able to tolerate different skin types, damages, dryness, as well as the humidity of the finger surface.

2. Image quality improvement. By using image quality improvement, an optical improvement of the structures (ridges) on the scanned image can be achieved.

3. Image processing. Image processing means the preparatory phase for feature extraction and classification purposes.

4. Feature classification. Fact is that all fingerprints show certain global similarities, which allow for rough classification into three principal finger classes. However, classification is a rather difficult process both for algorithm-based decisions as well as for man-made decisions since some



fingerprints cannot be clearly allocated to a concrete finger class. Nowadays, pattern classification is only used in dactyloscopic systems, e.g. AFIS (Automated Fingerprint Identification System) of the Federal Office of Criminal Investigation (BKA). This method is not feasible for access systems.

5. Feature extraction. In this phase, the location of the minutiae (ridge bifurcations and ridge endings) in the fingerprint is detected and extracted. In practice, scanned fingerprint images show differing qualities. The algorithm performance is negatively influenced by a poor image quality.

6. Verification phase. In the verification phase two feature vectors are being compared. The algorithm performance strongly depends on the quality (significance) of the extracted minutiae and on the comparison process.

Below, we will describe in more detail the fingerprint scanning, feature classification, and fingerprint comparison processes.

2.4.2 Fingerprint Scanning

Depending on whether the scanning process is carried out on- or off-line, the fingerprint image can either be • a color image, e.g. on paper or • an image of a life finger obtained through a sensor In case of a color print, rolling the finger on a surface generates the image of the ridges, e.g. on paper after that the finger is moistened with ink. An example of such rolled fingerprint images can be seen in Figure 2.5. By evenly rolling the finger from one side of the nail to the other, all line-related data is reliably recorded in the image. Afterwards, these images can be scanned or electronically photographed. In dactyloscopy, this method has already been used for well over 100 years. Thus, since a complete "overall imaging" of the finger is done, in addition to a higher number of ridges and minutiae, "macro features" (i.e. delta and nucleus) are recorded as well. Even though they are part of each and every ridge pattern (with the exception of the "arch" pattern which does not have a delta), they are not always printed. A disadvantage of this method is a possible distortion, which may occur through pressing and rolling the finger while taking the fingerprint. Furthermore, a quality feedback is not possible which may lead to a decrease in the quality of the fingerprints. From the user’s point of view, this is an unpleasant and slow method. It is also unsuitable for partially automated access controls.

Figure 2.5: Color image

The term life image of a finger is a comprehensive term for images that are obtained directly by placing the finger on a suitable sensor. There are a vast number of various methods, which can be used for scanning ridges. They include: • Optical sensors, • Electrical field sensors, • Polymer TFT sensors (TFT – Thin Film Transistor), • Thermal sensors,



• Capacitive sensors, • Contactless 3D-sensors, and • Ultrasound sensors. A biometric sensor is the hardware component of a biometric system, which initially supplies biometric measurements. Depending on the biometric method in use, there are different kinds of sensors. Optical sensors use light for obtaining fingerprint images – confer Figure 2.6. Electrical field sensors measure local variations of the electrical field, which is generated on the finger surface relief upon the emission of a small electrical signal. Polymer TFT sensors measure the light, which is emitted upon contact when the finger is laid on the polymer substrate. Thermal sensors register the thermal finger image. In capacitive sensors, the sensor and the finger surfaces together form a capacitor. The capacity thereof changes based on the skin relief (skin ridges and grooves) – confer Figure 2.6. These local changes are measured and thus represent the fingerprint. The above-mentioned sensors are used in connection with the data-processing module as on-line systems. They substitute the off-line method in which fingerprints are, for example, taken on paper before they are digitised later on. Image quality strongly depends on the "contrasts" that were achieved between the ridges and the adjacent grooves. Since there is a feedback to image-processing algorithms for on-line methods, it is relatively easy to immediately check the quality of fingerprint images that were just obtained. The life image is usually recorded by lightly placing the finger on the surface of the sensor. Since it is not so user-friendly, the finger's surface is only rolled in the context of AFIS-systems (as in the dactyloscopic method). Of course, in this case, only such ridges can be captured that are recorded as a result of being directly in contact with the sensor surface. Hence, compared to rolled fingerprint images, the life image generates the image of a smaller section of the finger's surface but, in addition, it might also have smaller distortions of the image.

Figure 2.6: Capacitive sensor, optical sensor

Currently, the most frequently used life image technology is the optical method. Upon placing the finger on the sensor's glass pane (prism), the elevations of the papillary lines are in contact with the glass; the grooves, on the other hand, are not in contact with it. Basically, the recording device consists of a light source (LED) and a CCD camera, both of which are located within the device on the other side of the glass pane. The light of the LED illuminates the glass at a certain angle and the photo-element receives the reflected light. The course of the beam runs in such a way that the incoming light on the contact ridges is scattered as if on a mirror surface and then reflected back on the CCD camera. There where the grooves are behind the glass pane, the light passes through; these spots remain dark. An example for such a fingerprint image can be seen in Figure 2.7. Figure 2.8 shows the design of some types of sensors.



Figure 2.7: Life image of a finger

An ever re-occurring question is that of life recognition of a finger (temperature, fluorescence, pulse), the question being whether a scanner or sensor would accept an artificial finger. This question was already addressed in a study entitled BioIS [ZWI00]. The conclusions reached in this study have not changed significantly until today [CT02].

Figure 2.8: Types of sensors

2.4.3 Pattern Classification

The global pattern of papillary lines occurring in the central area of the tip of the finger constitutes a specific configuration, which is sufficient for a rough systematic classification. For fingerprint classification purposes, only a part of the entire image, called Pattern Area, is used. The Pattern Area is defined as the inner area, which is limited by two lines, so-called Type Lines. Two singular points are part of this central area of the fingerprint image (Confer Figure 2.9): (a) the delta (several of which may exist; only sample arches do not have deltas) and (b) the nucleus. Delta, which is sometimes also called "outer border" is usually located at the fringe of the fingerprint image. An image of papillary lines is called a delta if it is similar to the Greek capital letter delta. It is formed by two parting ridges or by a ridge bifurcation and a third ridge that is convex and coming from another direction. Some examples of a delta configuration are shown in Figure 2.10. It is rather hard to define the nucleus of an individual fingerprint due to vast variations in the curving of the inner lines. Therefore, a specific point is simply chosen as the nucleus as though it was the center of the corresponding pattern. Figure 2.11 shows some examples of a nucleus configuration. Another important quantitative factor in classifying images is the number of lines. This means the number of lines that touch or cross the imaginary connection between the nucleus and the delta. Due to the great complexity of various line configurations, it is often difficult to clearly determine the number of lines. Figure 2.12 shows three simple examples for the number of lines.



Figure 2.9: Type lines

Figure 2.10: Delta configurations

Figure 2.11: Nucleus configurations

Figure 2.12: Examples of the number of lines

According to the definitions given above, fingerprint categories can be described as follows (pursuant to the Henry classification system [HEN03]): In loops, one or more ridges enter into the central area, they form a curve, touch or cross the imaginary lines between the delta and the nucleus and return to the same side from which they came. There are three decisive characteristics for classifying lines as a loop: (a) at least one suitably curved papillary line, (b) a delta, and (c) a number of lines other than zero. Depending on the orientation of the line's curve, a differentiation is made between right (clockwise) and left (anticlockwise) loops. Approximately 60 to 65 % of human fingerprints belong into this category. Whorls have at least two deltas. In their nucleus, ridges form a twist. Even though this definition is very general, it expresses the main characteristic of this category. Whorls can be split up into further



categories: (a) flat whorls, (b) whorls with a medium slant, (c) double whorls, and (d) random whorls. About 30 to 35 % of all fingerprints belong into this category. Arches are a rather special type of fingerprint. Less than 5 % of all fingerprints belong into this category. Arches can be split up into two categories: (a) flat arches and (b) high arches. In flat arches, the ridges enter at the side, form moderate and nearly parallel waves in the center and exit on the opposite side. In high arches, the wave is stronger in the middle. The route of all lines is no longer parallel and part of the lines seemingly exerts pressure from below.

Figure 2.13: Flat arch, left loop, right loop, high arch, and whorl

Obviously, due to the vast variation in the spectrum of fingerprints, the classification is always a big problem both for experts as well as for automatic systems. The allocation into categories is a very complex task. Dactyloscopy experts need a lot of experience in order to do their work reliably. Figure 2.13 shows some examples for individual categories. Figure 2.14 demonstrates examples of fingerprint images, which are very difficult to classify.

Figure 2.14: Left loop, high arch

2.4.4 Fingerprint Image Comparison

Data about the fingerprint category and further global characteristics, such as the number and position of the centers, deltas, and ridges, does provide enough information for a certain differentiation of fingerprints. However, the true individuality of fingerprints is determined by the anatomic characteristics of the ridges (minutiae) and their respective orientation. Whether they can be recorded in their entirety depends on the conditions when the fingerprint was taken as well as on its quality. The most frequently occurring minutiae are • Ridge ending and • Ridge bifurcation. Ridge ending defines the end of a line, while ridge bifurcation is defined as a point in the ridge where the line is separated into two branches. Minutiae are usually stable and robust with regard to conditions occurring during the scanning process. Figure 2.15 shows some examples. Minutiae can be characterized by their type, by x- and y-coordinates in a coordinate system, and by their direction. Figure 2.16 shows the directions.



Figure 2.15: Ridge ending, simple ridge bifurcation, twofold ridge bifurcation, threefold ridge bifurcation, simple whorl, twofold whorl, and side contact; Hook, point, interval, X-line, simple bridge, twofold bridge, and continuous line

If two fingerprints belong into the same category and have a certain number of identical minutiae, it is quite safe to say that they come from the same finger.

(x,y

(x,yα

α

X

Y

Ridge ending Bifurcation

Figure 2.16: Ridge ending and ridge bifurcation

The general definition for the identicalness of any two fingerprint images consists of four criteria and says: • The general pattern configuration has to be identical, • The minutiae have to be qualitatively identical (qualitative factor), • The quantitative factor says that a certain number of minutiae must be found (in Germany it is

12), and • There has to be a mutual minutiae relationship specifying that corresponding minutiae must have

a mutual relationship. In practice, a large number of complex identification protocols for fingerprint image comparisons have been proposed. These protocols are derived from the traditional dactyloscopic methodology and prescribe an exact procedure for trained specialists.

Even though various protocols differ in the process flow of the comparison procedure and the definition of the decision, the basic steps remain the same. Typically, comparison is done in an iterative three-phase-process. It is hardest to compare two fingerprints that have similar feature configurations. If, however, both fingerprints are totally different as far as their feature configuration is concerned, it is impossible that these images are from the same finger. In the next step, significant minutiae are examined, the central area is located, and the minutiae are compared with each other. Afterwards, the decisive comparison of the minutiae is carried out where all minutiae of the fingerprints are compared with each other. A decision is made based on identified pairs and their configuration. Due to variations in fingerprint qualities, not all points are always clear or defined with the same quality. In such cases, experts use their discretion and experience in deciding whether images are identical or not. For instance, ridge bifurcations could be identified as ridge endings if little pressure was exerted in taking the fingerprint. Obviously, the experience of the experts always plays a certain key role when comparing fingerprints. As an example, Figure 2.17 shows the comparison of 18 such minutiae.



Figure 2.17: Dactyloscopic comparison with 18 corresponding minutiae

2.4.5 Image of the Fingerprint Identification Procedure

In this section, the individual steps of the application "Fingerprint Image Recognition by Comparing Minutiae" are documented in pictures (Confer Figure 2.18).



Image quality improvementImage processing

Thinning Feature extraction

Direction field determination Fingeprint scanning

Figure 2.18: Minutiae-based fingerprint identification procedure



3 Evaluation of Biometric Systems

3.1 Description of the Evaluation Criteria

Different types of error rates are used as metrics for the operative capability of biometric authentication systems in general and for fingerprint image recognition systems in particular. The result of a comparison in the feature matcher within a fingerprint image recognition system is called Matching Score "s". It measures the similarity between the fingerprint image and the stored template. The closer s approaches 1 (if normalized between in the range [0,1]), the more likely it is that both fingerprints originate from the same finger. On the other hand, if s is near 0, it will be quite probable that both fingerprints are from different fingers. The decision of the system is determined by threshold T, i.e. if s passed the threshold, the fingerprints are regarded as being of the same finger (Matching Pair). If s is below the threshold, the fingerprints are regarded as being different (Non-Matching Pair). In connection with this, two erroneous decisions, i.e. two kinds of mistakes, can be made by biometric systems. • False Match – Two fingerprint images of different fingers are categorized as being identical. • False Non-Match – Two fingerprints of the same finger are categorized as being different. These two mistakes are often referred to as False Acceptance and False Rejection. In order to provide a clear understanding of the different kinds of errors that can occur, they shall be defined below.

3.1.1 Types of Errors

False Acceptance Rate (FAR) The FAR is the probability that a biometric system falsely recognizes different characteristics as identical, thus failing to reject, for example, a potential intruder. Definition:

Number of comparisons of different fingers resulting in a match FAR= Total number of comparisons of different fingers

False Rejection Rate (FRR) The False Rejection Rate (FRR) is the probability that a biometric system falsely recognizes identical characteristics as being different, thus, for example refusing to accept an authorized person. Definition:

Number of comparisons of the same fingers resulting in a non-match FRR= Total number of comparisons of the same fingers

False Match Rate (FMR) The False Match Rate (FMR) indicates the proportion of persons who, in the characteristics comparison, were falsely accepted. Those attempts that were previously rejected (Failure To Acquire, FTA) due to a low quality (e.g. of the image) are, in contrast to FAR, not taken into consideration. Please note that it depends on the application whether a falsely accepted characteristic contributes to increasing the FAR or FRR.



False Non-Match Rate (FNMR) The False Non-Match Rate (FNMR) indicates the proportion of persons who, when comparing characteristics, were falsely not accepted. Those attempts that were previously rejected (Failure to Acquire, FTA) due to a low quality (e.g. of the image) are, in contrast to FRR, not taken into consideration. Again, it depends on the application whether a falsely non-accepted characteristic contributes to increasing the FRR or FAR. In contrast to the FAR and FRR, which are often used metrics in literature, the FMR and FNMR are calculated by the enrolled template through a number of comparisons. In contrast to it, the FAR and FRR are calculated via transactions and include, for example, the Failure to Acquire (FTA, confer below) rates as well. Figure 3.1.1 outlines how the FMR and FNMR are calculated. The definitions of the error rates result from the probability densities for the comparison of different and identical fingerprints with regard to the threshold T:

( )∫=1

|)(T

uu dsHspTFMR ( )∫=T

gg dsHspTFNMR0

|)(

with the • Decision threshold T, • Statement "different" Hu (enrolled fingerprint and template come from different fingers), • Statement "identical" Hg (enrolled fingerprint and template come from the same finger), • Probability density p which fulfils the hypothesis in brackets, and • Matching Score s.

p

s

Person Person

T

FNMR(

T)

FMR(

T)

10

Hu Hg

Threshold

Different features

Same feature

Figure 3.1.1: FMR and FNMR

Thus, the two error rates depend both on the probability densities pu(s | Hu) and pg(s | Hg), which characterize the system. They are a function of threshold T. Equal Error Rate (EER). The EER is defined by the condition FNMR (T) = FMR (T). In practice, the probability densities are discrete functions. Hence, the EER cannot be determined exactly. In contrast, an EER range can be established in which error rates match. As a result, if threshold T of the system is set accordingly, the same number of people are falsely accepted and falsely rejected. Furthermore, depending on the application, fixing threshold T in such a way that different error rates are generated, can be useful.



0.00.10.20.30.40.50.60.70.80.91.0

Threshold T

FMR FNMR EER

0

Figure 3.1.2: FAR, FRR, and EER in dependence on threshold T

ZeroFNMR is the lowest limit of FMR, i.e. FNMR = 0. ZeroFMR is the lowest limit of FNMR, i.e. FMR = 0.

EER

0 1Threshold T

Erro

r

FMR FNMR

ZeroFMR ZeroFNMR

Figure 3.1.3: ZeroFMR, ZeroFNMR, and EER

The Failure To Acquire Rate (FTA) reflects the frequency at which a fingerprint image cannot be acquired by the sensor in automatic mode. This means the scanning of the fingerprint was rejected even though the finger was placed on the sensor. The higher the measured value, the less the sensor is suited for acquiring the fingerprint. In this sense, the error rate is a parameter for evaluating the sensor. The Failure To Enroll Rate (FTE) indicates the percentage of identities which cannot be enrolled by the biometric recognition system. FTE rates occur often in connection with systems which, by checking the fingerprint image quality, decide whether or not a template will be generated. This means that low quality fingerprint images will not be enrolled in the system. In this sense, FTE is a parameter evaluating the capability of the algorithm to process low quality fingerprint images. The Failure To Match Rate (FTM) indicates the percentage of enrolled fingerprints which can neither be matched nor generally processed with the stored biometric templates. This shows the incapability of the system to make a decision, i.e. unlike in case of false matches there is no result in which a wrong decision could be made.



3.1.2 Objective Comparison of Fingerprint Systems

Both FMR and FNMR depend on threshold T that was set and thus they are a function of T. For example, if the threshold is moved to the left (i.e. reduced), in order to increase the tolerance of the whole system, FNMR decreases and FMR rises accordingly. The system's performance in different operating points (threshold 7) can be shown in a ROC (Receiver Operating Characteristic) curve. This curve plots the FRR versus FAR thus eliminating the graph's dependence on threshold T. ROC curves are the standard approach for evaluating the performance of pattern recognition systems. These curves provide for objective comparisons in decision systems. Hence, they can be applied when comparing biometric systems in general and fingerprint recognition systems in particular. ROC curves either show the detecting rate (1 - FRR) or they show FRR as an FAR function. The graph chosen here shows FRR = f(FAR).

0.001

0.01

0.1

1 1 0.1 0.01 0.001

FAR

FRR

EER

T falling

Figure 3.1.4: ROC -curve

ROC curves offer the possibility of determining different operating points. A possible operating point is, e.g. the operating line for identical errors. The operating point of EER is determined by the (point of) intersection of the ROC curve and the straight line FRR = FAR. Linking Various Error Rates for the Purpose of Objective Comparisons

If we look at a biometric recognition system from the outside as a Black Box5, it does not matter where the FAR and FRR error rates come from. They consist of (1) errors resulting from the acquisition of images (FTA), (2) errors from enrolling fingerprints (FTE), and (3) FNMR and FMR errors resulting from the actual comparison of fingerprints: • The FTA rate describes the percentage of fingerprints that could not be aquired. A higher FTA

increases FRR and, on the other hand, decreases FAR. Hence, the portion of fingerprints that could be aquired is . ( )FTA−1

• The FTE rate describes the percentage of fingerprints, which could not be enrolled by their respective algorithms. Higher FTEs increase FRR and, consequently, reduce FAR. Hence, the portion of fingerprints that could be enrolled is ( )FTE−1 .

Consequently, this results in the following combined error rates: • : This is the proportion of fingerprints, which could be acquired but not( ) FTEFTA ×−1

enrolled.

5 The result is tested via the user interface without knowledge of Interna.



• : This is the proportion of fingerprints, which could be both acquired and( ) ( FTEFTA −×− 11 ) enrolled.

Together, various error rates have been combined into a verification system as follows:

( ) ( ) ( )( ) ( ) ( ) TFNMRFTEFTAFTEFTAFTATFRR

TFMRFTEFTATFAR×−×−+×−+=

×−( )

×−=111)(

)(11

The boundary conditions for the matching rates (Confer Figure 3.1.1) are:

1)1(,0)0(0)1(,1)0(

====

FNMRFNMRFMRFMR

3.2 Experimental Determination of the ROC curves

3.2.1 Determination of the Probability Density Functions

In practice, probability density functions are generated in experiments by comparing identical and differing fingerprints if no reliable theoretical model for probability densities can be drawn up. The result of the comparison is the Matching Score s. In Figure 3.2.1, the left-hand side function is a corresponding function for comparing different fingers, while the curve on the right-hand side is a result of comparing fingerprint images of identical fingers. Generally, the two curves overlap resulting in the error rates described in Section 3.1.1.

Figure 3.2.1: Error rates for the comparison of identical and differing

3.2.2 Calculation of FNMR(T) and FMR(T)

The values of FNMR(T) and FMR(T) are generated respectively for each threshold T (Confer Figure 3.2.1. Threshold T is marked).



In order to determine FNMR(T), fingerprint images of the same finger have to be compared. This means a database of different fingerprint images of the same finger is generated. By using different thresholds T, the FNMR error rate can be determined as a function of T in an experiment. The number of comparisons NFNMR is the result of the number of comparisons per finger NMF and the number of fingers in the database NFDB (NAF = enrolled number of images per finger):

FDBAFAF

N

iFDBAFFDBMFFNMR N

NNNiNNNN

AF

*2

)1(*)(

1

1

−=−=×= ∑

−

=

In order to determine FMR(T), fingerprints of different fingers have to be compared. This means a database of different fingers has to be set-up. Analogously to the above-mentioned case, the FMR error rate can be determined as a function of T when varying the T threshold in an experiment. In this case, the number of comparisons NFMR is determined by the number of different fingers in the database NFDB, which are matched up in pairs.

2)1(

)(1

1

−=−= ∑

−

=

FDBFDBN

iFDBFMR

NNiNN

FDB

Generally, the threshold T varies between 0 and 1. The values of FNMR(T) and FMR(T) are placed on a graph toward the threshold T (Confer Figure 3.2.2).

Threshold T10

FMR(

T )1

Mat

chin

g Sc

ore

s

FNMR(

T)

EER

Figure 3.2.2: FNMR(T) and FMR(T)

3.2.3 Determination of the ROC curves

For determining the ROC curve, a pair of values (FRR(T), FAR (T)) is defined for each value of threshold T. Thus parameter T is eliminated and the performance of the biometric system can be shown as being independent of the threshold. This is particularly important with regard to objectively comparing different systems whose error rates, being functions of threshold T, may have totally different paths. If FAR(T) is plotted on the x-axis and FRR(T) on the y-axis of the ROC curve, then the path of the ROC curve, at an increasing threshold T, starts in the right bottom corner and proceeds to the left top corner.



0.001

0.01

0.1

1 1 0.1 0.01 0.001

FAR

FRR

T rising

Figure 3.2.3: Example of a ROC curve

3.3 Police-Related Application Scenarios of Biometrics Systems and their Requirements regarding Error Rates

1. Using different biometric recognition systems for verification purposes, e.g. as part of border controls (country entry/exit controls, 1 to 1 comparison).

FRR > FAR (FAR ≅ 0)

2. In addition to analyzing video tapes by using biometric recognition methods, these systems can effectively support video surveillance in high-crime areas and in situations (such as not approved demonstrations or tumults at sports events) in an automated way and shortly after their occurrence. The necessary precondition is that face recognition systems are able to recognize faces of moving people even from inadequate viewing directions and to compare them on-line with existing databases.

FRR < FAR

3. Biometric comparison of face images with phantom images (1 to n). FRR < FAR

4. Mobile use of fingerprint scanners (local identity check-ups, 1 to n). FRR < FAR

5. Forensic processing using biometric fingerprint scans and AFIS-comparisons (1 to n). FRR < FAR

6. Identification of repeat offenders using biometric characteristics such as voice and walk (e.g. video tapes of bank robberies, 1 to n).

FRR < FAR

While in the first scenario, there usually is no further examination of the result of biometric matching, the results of biometric systems in scenarios nos. 2 to 6 simply serve as approaches for investigations or hints which would definitely have to be verified, e.g. by forensic experts.



FRR

FAR

1

1

FAR = FRRFA

R =

0 FRR > FAR

FAR > FRR

Figure 3.3.1: ROC curves and corresponding FAR / FRR data

Figure 3.3.1 shows the operating regions and lines of the systems for various scenarios. On the left-hand side (toward the y-axis) of the green line there is the area for systems of border and passport controls for verification purposes. On the other side of the green line (toward the x-axis) lies the area for identification systems.



4 Investigations with Test Persons

4.1 Inclusion of the Database

Principally, there are two possibilities to examine the influence of sensors on the recognition rates. 1 Several enrollment procedures are carried out and the respective fingerprint images acquired. The

conditions surrounding the procedure are considered to be constant. However, these conditions will usually vary when comparing two procedures. This implies that the robustness of the sensors toward environmental influences is thus taken implicitly into account. Furthermore, physical parameters of the environment can directly be adjusted prior to acquiring the respective data.

2 Upon acquiring a data record, the influence of environmental conditions on the sensors and the respective fingerprint images can be simulated using image processing operations.

The two procedures mentioned above have the following advantages and disadvantages: • Comparing individual sensors is problematic since keeping external influences constant while

recording individual test series is difficult. However, if, prior to storing the original image, the systems carry out internal image processing, this is the only possibility of explicitly taking this influence into consideration.

• If, as mentioned in No. 1, systems do carry out internal image processing, then image processing operations cannot be applied for simulating the influence on fingerprint images since these operations are carried out only after the images have been stored and hence do not take internal processing of the sensor into consideration. On the other hand, different operations on the fingerprint images offer the possibility of generating a large database of fingerprint images. Changes therein are defined by image processing operations and can be reproduced at any time.

Within the BioFinger Project, the database is established according to the first procedure since the internal image processing operations of various fingerprint image scanners were to be incorporated into the test.

4.1.1 Sensors and Algorithms

For each scanner a database with fingerprint images is recorded. This means that for NS sensors, which participate in the test, databases are defined as DBi, i=1,.., NS.



Technology Name of the sensor Image size Resolution pressure-sensitive

BACU-100 (Hitachi/BMF) 256−384 440 dpi

capacitive TouchChip™ (STMicroelectronics) 256−360 508 dpi capacitive ID Mouse (Infineon) 224−288 513 dpi capacitive AES4000 (AuthenTec) 193−193 250 dpi

optical TFS 050 (TBS) 576−744 500 dpi optical TST BiRD IIi (TST) 320−384 500 dpi optical ACCO USB (Heimann) 376−472 500 dpi optical FX 2000 (Biometrika SRL) 316−376 569 dpi optical LS2™ / F (Heimann) 740−580 500 dpi optical MorphoSmart™ MSO100 (Sagem) 416−416 500 dpi thermal FingerChip™ AT77C101B (Atmel) 280−320 280 dpi

paper BKA data record 780−780 500 dpi

Table 4.1.1: Tested sensors

4.1.2 Description of the Sensors

• Product name: USB Fingerprint Reader BACU-100 • Description: access system • Figure:

• Sensor dimensions – thickness [mm]: 1.9 • Sensor dimensions – width [mm]: 24.4 • Sensor dimensions – height [mm]: 35.9 • Sensor dimensions – weight [g]: not available • Sensor technology: pressure-sensitive • Resolution [dpi]: 440 • Life expectancy: > 1.2 million fingertips • Sensor dimensions – thickness [mm]: 10.8 • Sensor dimensions – width [mm]: 70.3 • Sensor dimensions – height [mm]: 100.9 • Sensor dimensions – weight [g]: not available



• Type of connection: USB • Algorithm: not available • Minutiae: not available • FAR: 0.01 % • FRR: 0.001 % • Template – format: not available • Template – size [bytes]: 400,000 • SDK available (price): not available • Storage of the original image: not available • Storage of extracted minutiae: not available • Input of modified images: not available • Input of modified minutiae: not available • Protocols: not available • Operating system: MS Windows, Linux, Mac • Installation: setup from the CD • Life recognition: no • Rolled-off fingerprints: no • Manufacturer: BMF Corporation, www.bm-f.com • Price: 90 € • Product name: TouchChip™ Fingerprint Sensor TCS1CD • Description: sensor • Figure:

• • Sensor dimensions – thickness [mm]: 3.5 • Sensor dimensions – width [mm]: 20.4 • Sensor dimensions – height [mm]: 27 • Sensor dimensions – weight [g]: not available • Sensor technology: capacitive • Resolution [dpi]: 508 • Life expectancy: not available • Sensor dimensions – thickness [mm]: not available • Sensor dimensions – width [mm]: not available • Sensor dimensions – height [mm]: not available • Sensor dimensions – weight [g]: not available



• Type of connection: 20-pin Flex connection • Algorithm: not available • Minutiae: not available • FAR: not available • FRR: not available • Template – format: not available • Template – size [bytes]: not available • SDK available (price): yes (not available) • Storage of the original image: not available • Storage of extracted minutiae: not available • Input of modified images: not available • Input of modified minutiae: not available • Protocols: not available • Operating system: not available • Installation: not available • Life recognition: not available • Rolled-off fingerprints: not available • Manufacturer: STMicroelectronics GmbH, www.stm.com • Price: not available • Product name: ID Mouse Professional • Description: access system • Figure:

• Sensor dimensions – thickness [mm]: 1.5 • Sensor dimensions – width [mm]: 18 • Sensor dimensions – height [mm]: 21 • Sensor dimensions – weight [g]: not available • Sensor technology: capacitive • Resolution [dpi]: 513 • Life expectancy: > 100,000 fingertips • Sensor dimensions – thickness [mm]: 127 • Sensor dimensions – width [mm]: 65 • Sensor dimensions – height [mm]: 30 • Sensor dimensions – weight [g]: not available



• Type of connection: USB • Algorithm: feature / image comparison • Minutiae: minutiae + pattern • FAR: < 10-7 • FRR: 0.1 % to 10 % (dependent on algorithm) • Template – format: DIN V 66400 • Template – size [bytes]: 100 – 200,000 • SDK available (price): yes (2,500 €) • Storage of the original image: yes • Storage of extracted minutiae: yes • Input of modified images: yes • Input of modified minutiae: yes • Protocols: not available • Operating system: MS Windows 98 / NT / ME / 2000 / XP • Installation: setup from CD or ZIP (from the Internet) • Life recognition: no • Rolled-off fingerprints: no • Manufacturer:Siemens AG – ICM RDC IS BIO, www.siemens.com/biometrie • Price: 101.75 €



• Product name: AES4000 TEK for USB • Description: access system • Figure:

• Sensor dimensions – thickness [mm]: 1.4 • Sensor dimensions – width [mm]: 20 • Sensor dimensions – height [mm]: 20 • Sensor dimensions – weight [g]: not available • Sensor technology: capacitive • Resolution [dpi]: 250 • Life expectancy: not available • Sensor dimensions – thickness [mm]: not available • Sensor dimensions – width [mm]: not available • Sensor dimensions – height [mm]: not available • Sensor dimensions – weight [g]: not available • Type of connection: USB • Algorithm: not available • Minutiae: not available • FAR: not available • FRR: not available • Template – format: not available • Template – size [bytes]: not available • SDK available (price): not available • Storage of the original image: not available • Storage of extracted minutiae: not available • Input of modified images: not available • Input of modified minutiae: not available • Protocols: not available • Operating system: MS Windows 98 / ME / XP • Installation: not available • Life recognition: not available • Rolled-off fingerprints: not available • Manufacturer: AuthenTec Inc., www.authentec.com • Price: $ 149.00



• Product name: TFS 050 • Description: access system • Figure:

• Sensor dimensions – thickness [mm]: 30 • Sensor dimensions – width [mm]: 40 • Sensor dimensions – height [mm]: 50 • Sensor dimensions – weight [g]: 180 • Sensor technology: optical • Resolution [dpi]: up to 1,000 • Life expectancy: > 5 years • Sensor dimensions – thickness [mm]: 30 • Sensor dimensions – width [mm]: 40 • Sensor dimensions – height [mm]: 50 • Sensor dimensions – weight [g]: 180 • Type of connection: Video, Ethernet, FireWire, optional USB 2.0 • Algorithm: Sagem / Morpho • Minutiae: minutiae • FAR: 0.01 % • FRR: 0.1 % • Template – format: binary • Template – size [bytes]: 512 • SDK available (price): yes (1,400 €) • Storage of the original image: yes • Storage of extracted minutiae: yes • Input of modified images: yes • Input of modified minutiae: yes • Protocols: serial • Operating system: MS Windows • Installation: setup from CD • Life recognition: yes • Rolled-off fingerprints: yes (with TRS 050) • Manufacturer: Touchless Biometric Systems AG, www.tbsinc.de • Price: from 300 €



• Product name: TST BiRD IIi • Description: contactless fingerprint access system • Figure:

• Sensor dimensions – thickness [mm]: not available • Sensor dimensions – width [mm]: not available • Sensor dimensions – height [mm]: not available • Sensor dimensions – weight [g]: not available • Sensor technology: optical • Resolution [dpi]: 500 • Life expectancy: not available • Sensor dimensions – thickness [mm]: 164 • Sensor dimensions – width [mm]: 100 • Sensor dimensions – height [mm]: 185 • Sensor dimensions – weight [g]: 530 • Type of connection: USB, Ethernet • Algorithm: feature comparison • Minutiae: not available • FAR: 0.05 % • FRR: 5 % • Template – format: not available • Template – size [bytes]: 500 • SDK available (price): yes • Storage of the original image: yes • Storage of extracted minutiae: yes • Input of modified images: no • Input of modified minutiae: no • Protocols: TCP/IP • Operating system: MS Windows 98 / ME / 2000 / XP • Installation: Installationswizard • Life recognition: yes • Rolled-off fingerprints: no • Manufacturer: TST – Touchless Sensor Technology AG, www.tst-ag.de • Price: 2,500 €



• Product name: ACCO USB • Description: dactyloscopic system • Figure:

• Sensor dimensions – thickness [mm]: not available • Sensor dimensions – width [mm]: 19 • Sensor dimensions – height [mm]: 24 • Sensor dimensions – weight [g]: not available • Sensor technology: optical • Resolution [dpi]: 500 • Life expectancy: MTBF > 20,000 hours • Sensor dimensions – thickness [mm]: 120 • Sensor dimensions – width [mm]: 60 • Sensor dimensions – height [mm]: 53 • Sensor dimensions – weight [g]: 550 • Type of connection: USB • Algorithm: fingerprints are transmitted to AFIS • Minutiae: not available • FAR: not available • FRR: not available • Template – format: not available • Template – size [bytes]: not available • SDK available (price): no • Storage of the original image: yes • Storage of extracted minutiae: not available • Input of modified images: no • Input of modified minutiae: not available • Protocols: ANSI / NIST-ITL 1-2000 and raw images in constant image formats • Operating system: MS Windows 2000 Professional / XP Professional • Installation: setup from CD (MSI Installer) • Life recognition: no • Rolled-off fingerprints: no • Manufacturer: Heimann Biometric Systems GmbH, www.hbs-jena.com • Price: project-specific



• Product name: FX 2000 • Description: access system • Figure:

• Sensor dimensions – thickness [mm]: 73 • Sensor dimensions – width [mm]: 23 • Sensor dimensions – height [mm]: 55 • Sensor dimensions – weight [g]: 310 • Sensor technology: optical • Resolution [dpi]: 569 • Life expectancy: > 5 years • Sensor dimensions – thickness [mm]: 125 • Sensor dimensions – width [mm]: 70 • Sensor dimensions – height [mm]: 60 • Sensor dimensions – weight [g]: 380 • Type of connection: USB, parallel • Algorithm: feature comparison • Minutiae: ridge bifurcation, ridge ending • FAR: 0.165 % • FRR: 0.249 % • Template – format: own • Template – size [bytes]: 2,000 • SDK available (price): yes (1,022 €) • Storage of the original image: yes • Storage of extracted minutiae: yes • Input of modified images: yes • Input of modified minutiae: yes • Protocols: 128-bit symmetric code • Operating system: MS Windows 95 / 98 / NT / ME / 2000, Linux • Installation: Plug & Play • Life recognition: is being developed • Rolled-off fingerprints: yes • Manufacturer: Biometrika SRL, www.biometrika.it/eng • Price: 185 € (or 208 € with smart card reader)



• Product name: LS2™ / F • Description: dactyloscopic system • Figure:

• Sensor dimensions – thickness [mm]: not available • Sensor dimensions – width [mm]: not available • Sensor dimensions – height [mm]: not available • Sensor dimensions – weight [g]: not available • Sensor technology: optical • Resolution [dpi]: 500 • Life expectancy: MTBF > 20,000 hours • Sensor dimensions – thickness [mm]: 290 • Sensor dimensions – width [mm]: 173 • Sensor dimensions – height [mm]: 130 • Sensor dimensions – weight [g]: 4000 • Type of connection: FireWire (IEEE 1394) • Algorithm: fingerprints are transmitted to AFIS • Minutiae: not available • FAR: not available • FRR: not available • Template – format: not available • Template – size [bytes]: not available • SDK available (price): no • Storage of the original image: yes • Storage of extracted minutiae: not available • Input of modified images: no • Input of modified minutiae: not available • Protocols: ANSI / NIST-ITL 1-2000 • Operating system: MS Windows 2000 Professional / XP Professional • Installation: will be installed by HBS-Service-staff members (in situ) • Life recognition: no • Rolled-off fingerprints: yes • Manufacturer: Heimann Biometric Systems GmbH, www.hbs-jena.com • Price: project-specific



• Product name: MorphoSmart™ MSO 100 • Description: access system • Figure:

• Sensor dimensions – thickness [mm]: not available • Sensor dimensions – width [mm]: not available • Sensor dimensions – height [mm]: not available • Sensor dimensions – weight [g]: not available • Sensor technology: optical • Resolution [dpi]: 500 • Life expectancy: not available • Sensor dimensions – thickness [mm]: not available • Sensor dimensions – width [mm]: not available • Sensor dimensions – height [mm]: not available • Sensor dimensions – weight [g]: not available • Type of connection: USB • Algorithm: not available • Minutiae: not available • FAR: not available • FRR: not available • Template – format: not available • Template – size [bytes]: not available • SDK available (price): yes (not available) • Storage of the original image: not available • Storage of extracted minutiae: not available • Input of modified images: not available • Input of modified minutiae: not available • Protocols: not available • Operating system: MS Windows 2000 • Installation: not available • Life recognition: not available • Rolled-off fingerprints: not available • Manufacturer: Sagem SA, www.sagem.fr/en • Price: not available



• Product name: FingerChip™ AT77C101B • Description: sensor • Figure:

• Sensor dimensions – thickness [mm]: 2.33 • Sensor dimensions – width [mm]: 9 • Sensor dimensions – height [mm]: 26.6 • Sensor dimensions – weight [g]: not available • Sensor technology: thermal • Resolution [dpi]: 280 • Life expectancy: > 1 million fingertips • Sensor dimensions – thickness [mm]: not available • Sensor dimensions – width [mm]: not available • Sensor dimensions – height [mm]: not available • Sensor dimensions – weight [g]: not available • Type of connection: not available • Algorithm: only hardware • Minutiae: not available • FAR: not available • FRR: not available • Template – format: not available • Template – size [bytes]: not available • SDK available (price): yes (not available) • Storage of the original image: not available • Storage of extracted minutiae: yes • Input of modified images: yes • Input of modified minutiae: yes • Protocols: not available • Operating system: not available • Installation: not available • Life recognition: not available • Rolled-off fingerprints: not available • Manufacturer: Atmel, www.atmel.com • Price: not available



The algorithms used are described accordingly with ALGi , i = 1,...,NALG.

Manufacturer of the algorithm Use in connection with the sensor Dermalog Only algorithm IKENDI AT77C101B (Atmel)

IDENCOM Hitachi/BMF IdentTechnologies AT77C101B (Atmel)

NEC Only algorithm Neurotechnologija TST

Siemens ID Mouse (Infineon), TouchChip (STMicroelectronics)

Table 4.1.2: Tested algorithms

4.1.3 Description of the Algorithms

Dermalog

• Extracted minutiae: ridge bifurcation, ridge ending, core, delta • Angle [step size]: yes • Minimum number of minutiae: 10 – 12 • Maximum number of minutiae: not available • Rotation ability: optional • Translation ability: optional • Ridge count possible: yes • Other aspects: among others, quality of the minutiae and neighborhood NEC

• Extracted minutiae: ridge bifurcation, ridge ending • Angle [step size]: yes • Minimum number of minutiae: 63 • Maximum number of minutiae: 191 • Rotation ability: ±30° • Translation ability: not available • Ridge count: yes • Other aspects: The fingerprint image is automatically categorized into zones of good and bad

quality. Siemens

• Extracted minutiae: ridge bifurcation, ridge ending • Angle [step size]: yes (not available) • Minimum number of minutiae: 3



• Maximum number of minutiae: not available • Rotation ability: optional (step 5°) • Translation ability: not available • Ridge count possible: yes • Other aspects: image surrounding of the minutiae papillary line width Neurotechnologija

• Extracted minutiae: ridge bifurcation, ridge ending • Angle [step size]: yes [1.40625°] • Minimum number of minutiae: not available • Maximum number of minutiae: 1024 • Rotation ability: 180° • Translation ability: not available • Ridge count possible: yes • Other aspects: curvature of the papillary line in the minutiae; singular points; block orientations IDENCOM

• Extracted minutiae: ridge bifurcation, ridge ending • Angle [step size]: yes (not available) • Minimum number of minutiae: not available • Maximum number of minutiae: 100 • Rotation ability: not available • Translation ability: not available • Ridge count possible: yes • Other aspects: grayscale and region information of the minutiae IKENDI

• Extracted minutiae: ridge ending and ridge bifurcations • Angle [step size]: yes [64 steps on 360 degrees] • Minimum number of minutiae: 5 • Maximum number of minutiae: 64 • Rotation ability: ±45° • Translation ability: 75% • Ridge count possible: not used • Other aspects: format complies with DIN-66400 CardCompact



IdentTechnologies

• Extracted minutiae: not available • Angle [step size]: not available [not available] • Minimum number of minutiae: not available • Maximum number of minutiae: not available • Rotation ability: not available • Translation ability: not available • Ridge count possible: yes • Other aspects: not available Test Data

A total of NP = 30 people took part in the test. The requirements for taking fingerprints are found below: • Each of the NP test persons had fingerprints taken for each of the NS test sensors. • Four fingerprints of both hands of each test person were taken (all fingers of one hand with the

exception of the small finger), NFP = 8. • The fingerprints of the test persons were taken in a total of ND = 3 sessions. • For each sensor NA = 3 takings were made per pass, i.e. there was a total of NAF = 9 images per

finger. Every single database for the various sensors thus contains NFDB = NP *NFP (e.g. NFDB = 240) different fingers. In each procedure, a total of NSession = NFDB *NA (e.g. NSession =720) fingerprints was acquired. This means that each database contains a total of NFADB = NSession *ND (e.g. NFA = 2160) fingerprints. Upon acquiring the fingerprint images, they were stored on the computer database and saved again after each session. In order to guarantee their unambiguous allocation, the image data was saved in different directories. It was possible to index them accurately according to the sensors, sessions, persons, and fingers. Hence, one directory contains NA = 3 images. If fingerprint images are missing in a directory, then the sensor could not acquire images of the finger of a certain person, which contributes to the Failure to Acquire (FTA) rate. Strategies to avoid errors when acquiring data: • Fingerprint scanners generate images in different resolutions. Thus, false allocations of images to

sensors can be recognized. • On the day when fingerprint images are made only a certain directory was opened for writing.

Thus, a correct image allocation to certain days was guaranteed. Other than sensors Nos. 1 to 11, fingerprint images of the database entitled "Sensor 13" were provided by the Federal Office of Criminal Investigation (BKA). This database will be described more in detail in Section 6.1. Types of Errors during Data Acquisition Naturally, when acquiring a large number of fingerprint images, a different kind of errors can occur. For example, if the acquisition software is not handled correctly, low-quality images are generated, for example if an image is acquired even though the finger has not been fully placed on the sensor. Following the enrollment procedure, this kind of errors can be detected if such images, which prove to have a Failure to Enroll (FTE) error, are examined by experienced analysts. Respective errors are documented in column No. 2 in the table below.



False allocations of fingerprints to persons were corrected and images with a very poor quality deleted. For this purpose, very low scores for matching identical fingers were examined (confer column No. 3 in the table below) and corresponding images sorted out. Error rates are described in Chapter 3.1.

Sensor Error caused during

data acquisition

False allocation or low

image quality

Remaining number

DB1 0t 0 2,160

DB2 18 4 2,138

DB3 7 55 2,098

DB4 2 12 2,146

DB5 3 0 2,083

DB6 7 4 2,149

DB7 3 11 2,146

DB8 2 0 2,158

DB9 0 1 2,135

DB10 6 0 2,154

DB11 1 0 2,135

Table 4.1.3: Corrected acquisition errors

The lower total number of fingerprint images for some of the sensors resulted from the fact that fingerprints of one or more persons could not be taken on the day of testing. Of greater importance for comparing different systems are the Failure To Acquire and Failure to Enroll rates (for further details refer to the section below), which can explain how sensors and algorithms respectively react to problematic fingerprint images.



FTE of Algorithms 1 – 7 Sensor Number FTA 1 2 3 4 5 6 7

DB1 2160 0 0 0 0.018055 0 0 0 0 DB2 2138 0.001403 0.005620 0 0.004215 0 0 0 0 DB3 2098 0.002859 0.056883 0 0.014340 0 0 0 0.008580 DB4 2146 0 0.007921 0 0.004659 0 0 0 0.010251 DB5 2083 0 0 0 0.000960 0 0 0 0 DB6 2149 0 0 0 0.090274 0 0 0 0 DB7 2146 0 0.031220 - 0.009319 0 - 0 0.064771 DB8 2158 0 0 0 0 0 0 0 0.063021 DB9 2135 0 0.009367 - 0.234192 0 - 0 0.008430 DB10 2154 0.012534 0 0 0.000470 0 0 0 0.003250 DB11 2114 0.009933 0 0 0 0 0 0 0 DB12 9504 0 0.002314 0 0 0.00021 0 0.00021 0.033670

Table 4.1.4: FTA- and FTE-rates

4.2 U1 – Influence of the Sensors on Verification

The aim of this investigation was to compare the performance of the sensors (as far as recognition performance is concerned). For this purpose, the recognition algorithm is kept constant and hence the performance of all tested sensors for this algorithm determined. By comparing the ROC curves it is possible to indicate the relative performance of the sensors with regard to the algorithm. Using one scanner, 72 fingerprint images of eight different fingers (four fingers per hand with the exception of the small finger) were taken of each test person, while acquiring nine fingerprints per finger respectively. This process was carried out for each of the 11 sensors that were part of the evaluation. In each of the eleven groups (one group per sensor), the following analyses for Matching Score Algorithms (MSA) can be carried out. At first, templates of nine fingerprint images are generated using feature extraction (ME) algorithms. After that, using the respective MSA algorithms, verifications within the group are carried out while keeping the algorithm constant.



Figure 4.2.1: Examination U1 - This process is carried out for each existing algorithm.

For each of the seven algorithms, eleven different densities pg(s | Hg) (according to the number of sensors) are generated. Consequently, probability densities of pu(s | Hu) are generated by comparing fingerprint images, which are not from the same finger. The various probability densities are determined for a fixed algorithm (again without combining different algorithms) for each of the eleven sensors. Using the probability densities pg(s | Hg), pu(s | Hu), which were determined in experiments, the respective error rates FNMR(T) and FMR(T) can be determined as a function of the threshold. This results in ROC curves, which indicate the influence of the sensors on the verification for fixed algorithms. The overall process is again illustrated in Figure 4.2.1.

4.3 U2 – Influence of Feature Extraction on Verification

For carrying out this kind of tests, the inter-operability of feature extraction and matching algorithms a necessary precondition. Unfortunately, in this survey, the templates of the algorithms, designed by different manufacturers, are not compatible. In this section we can therefore only deal with the planning of the examination scenario. Further information about the (lacking) inter-operability of the algorithms that were part of this study is to be found at the end of Section 4.3.



In order to evaluate the influence of various ME-procedures, the following procedure can be chosen: NB fingerprint images are stored in the database. According to the number of algorithms that are taking part in the test, MEj j=1,..,NME (Enroll_XXX*) feature extraction algorithms and MSAk k =1 ,..., NMSA (Match_XXX*) verification algorithms, which were provided by 7 different companies, are available. For the test, fingerprint image_X is selected from the database. The various templates for this image_X are calculated with all MEi (a total of 7 templates per image_X). These templates have to be stored in a reference format (e.g. in DIN V 66400 format) before they can be combined with other algorithms. Then, the MSAk algorithms are used for comparison purposes, which is the next step. It is important to evaluate all combinations of MEj and MSAk in order to check the quality of the characteristics that were generated. These combinations result in the respective FMR and FNMR values, which can be plotted in a ROC curve. For a better understanding of this procedure please refer to Figure 4.3.1.

ME1ME2

Database

Fig._01

Fig._02

Fig._NB

. .

.

Enroll_XXX

Template_ME1

Template_ME2

Template_ME3

Template_ME4

Template_ME5

Figure_X

Match_XXX

ME1ME2

ME1ME2

ME1ME2

ME1ME2

ROC1j

ROC2j

ROC3j

ROC4j

ROC5j

Figure 4.3.1: Influence of feature extraction on verification

The combination of the algorithms (MEj and MSAk) is important in order to be able to generate the ROC curves. For each MEj and for all possible MSAk, ROC curves are generated. Seven algorithms generate, for example, 49 ROC curves for all possible algorithm combinations. Thus, it is possible to compare all ROC curves with each other. In such a graph in which all ROC curves are combined, the "best" ROC curve can be determined. This “best“ curve should be near zero (the origin of the coordinate system). Hence, a statement can be made which combination of the MEj and MSAk algorithms is the “best“. Furthermore the “best“ ME algorithm and the “best“ MSA algorithm can be determined among all of the ROC curves. Furthermore it is important to note that the database remains constant for the duration of the examination. The different fingerprints must have been taken by the same sensor in order to avoid variations induced by different sensors. It remains to be seen whether the test can be realized in the future since some companies pursue a strictly confidential policy and hence do not release information about the structure of their template outside of the company. If the structure of the template is known, the question is whether other algorithms can work with this structure. The G-algorithm, for example, only extracts the ends of the valleys and the ends of the ridges. This data, however, is not compatible with the characteristics, which are used by the A-algorithm (which also uses ridge bifurcations).



The problem is even more complex because additional data is partly stored together with the fingerprint image. A-algorithm templates, for example, contain data about the region of the image around the features and, furthermore, the template is handled cryptographically. Thus, such a template absolutely cannot be used with any other MSAk functions. At this point, in order to show the problem of lacking inter-operability once again, the description of the content of templates made by different manufacturers is added: • Algorithm-A: Image quality, number of minutiae, data of minutiae are x- and y-coordinates,

angle, type (ridge bifurcation, ridge ending), image region of the minutiae and papillary line thickness. Combination of minutiae and information about image region. The templates are stored with double encoding.

• Algorithm-B: Image quality, number of minutiae, dimensions of the image; x- and y-positions of the minutiae, angle, grayscales of the minutiae and region information in x- and y-directions.

• Algorithm-C: Feature type with probabilities and alternatives, cores, deltas with coordinates, directions, position accuracy and quality; minutiae with coordinates, type, direction, quality and local ridge density as well as local curvature.

• Algorithm-D: Papillary line density, number of minutiae, x- and y-coordinates of the minutiae, angle and curvature in the point of the minutiae. Moreover, singular points of the fingerprint image and block orientations are extracted.

• Algorithm-E: Position and quality of the internal term; number of minutiae, minutiae (position, direction, type); good zones and unsuitable zones.

• Algorithm-F: Structure unknown. • Algorithm-G: Endings of valleys and ridges. A possibility to create inter-operability between different procedures is to insert standardized interfaces between the feature and verification algorithms. Some initial approaches in this direction can already be seen in the standardization of the DIN V 66400 format. For a detailed description of the DIN V 66400 format, confer [STR02].

4.4 U3 – Influence of the Algorithms (MSA) on Verification

The aim of this evaluation was to compare the performance (i.e. recognition performance) of the algorithms. For this purpose, the sensor type is kept constant and the performance of all algorithms that are part of the test is determined for that sensor. By comparing the ROC curves it is possible to indicate the relative performance of the algorithms with respect to the sensor. For each sensor, the fingerprints for a sensor are investigated for all algorithms in order to examine their influence on the verification. To study the influence of a specific algorithm the combination of feature extraction and comparison procedures of different manufacturers is not allowed (ME1 always together with MSA1 etc.). In this analysis, the influence of the algorithms on the performance, i.e. error rates, is measured by determination of the ROC curves for each of the sensors and the variation of different algorithms. For every sensor, this results in a host of seven ROC curves (one for each algorithm), with which the performance of the different algorithms for the respective sensors can be compared.



Fingerprints from person 12 Fingerprints not from person 12

MEj

MSAj

MEk

MSAk

10-3 10-2

S1S2

ROC

Figure 4.4.1: Evaluation of the influence of the algorithms on verification



Same fingerprints Different fingerprints

Enroll1

Match1

10-3

10-2

S1S2

ROC1

Same fingerprints Different fingerprints

Enroll5

Match5

10-3 10-2

S1S2

ROC5

Figure 4.4.2: Evaluation U3 – Sensor as parameter, the algorithm is varied

Artificial Change of the Template Moreover, a test approach which was not used here is outlined (the practice-oriented approach described above was chosen instead); however, it could be interesting for future analyses. It is a prerequisite in this approach as well as in the scenario described in Section 4.3 that the template format must be standardized. Once the template structure has been recognized, the data within the template can be changed. If it is changed artificially, the MSA can be tested, e.g. in order to determine the lowest number of features necessary for correct (successful) verification.



The following features can be changed in the template: • Number of minutiae, • Position of all minutiae (translation in x- or y-direction of the template), • Angle of all minutiae (rotation of the entire template), • Position of the individual minutiae (translation of the selected minutiae in x- or y-direction), • Angle of the individual minutiae (rotation of the gradient with selected minutiae).

4.5 U4 – Influence of the Sensors on Fingerprint Image Quality

In order to evaluate the quality of a fingerprint image, tests were carried out by using different metrics, which are described below.

4.5.1 Contrast

The contrast of an image is defined by relative, local intensity differences. Examples: 1. Weber contrast: dI / I 2. Michelson contrast: (Imax-Imin)/(Imax+Imin) The contrast for a complete image can be determined as the average of local intensity differences. Another approach is to examine intensity differences within individual, discrete basic intervals: the intensity difference between intensity i and intensity j I_(i,j) with i < j can be split up into "basic intensities" dI_(i,i+1), dI(i+1,i+2), ..., dI(j-1,j). In a high-contrast image it is possible that all intensity differences are evenly distributed. This will be the case if all pixels display the maximum intensity difference. If the distribution of the intensity differences differs from an even distribution, the contrast will decrease. Hence, the similarity with an even intensity distribution is a parameter in determining the contrast. If we consider the intensity differences as a vector in a vector space, which is set up by discrete basic intervals, then the image contrast can be compared to an even intensity distribution via the scalar product of the vector and the unit vector of an even intensity distribution.

4.5.2 Average Value of Grayscales

Histogram Here, the graphic representation of the occurrence of a given characteristic is called histogram. For example, the light colors (grayscales) of the picture elements (pixels) can be marked on the horizontal axis. The scaling of the horizontal axis is determined according to the brightness resolution in 256 channels. The 0 channel indicates the lowest brightness (black) and 255 equals the maximum brightness (white). In between are the remaining grayscales. When filling in an unweighted histogram, a "one" is added to the sum in the channel with the respective brightness for each pixel. If all pixels are examined, the result is a brightness or grayscale histogram of the fingerprint. The histogram can simply be calculated:

1210 −== LknnrHist k

k ...,,,,)(

Hist(rk) is the discrete histogram function; rk is the gray shading number k; nk is the number of pixels in channel rk; L is the number of grayscales, and n is the total number of pixels (e.g. 261,144 pixels at a resolution of 512×512). An example for a histogram is shown in Figure 4.5.1.



Grays

Freq

uenc

y

Figure 4.5.1: Example of a histogram

Normalization In order to compare histograms of fingerprint images of various sensors, normalization is required. The maximum (y-axis) has to be found in the histogram (in Figure 4.5.1 the maximum lies at the gray color 200). All values (for individual gray colors) are divided by this maximum value, i.e. all values will be element of the interval [0.1]. Then the maximum value is fixed at 1.0 and the remaining gray colors are scaled accordingly. Figure 4.5.2 shows a normalized histogram.

Grays

Nor

mal

ized

freq

uenc

y

Figure 4.5.2: Normalized histogram

Mean value The histogram in Figure 4.5.2 shows two peaks. The left-hand side peak corresponds to the papillary lines and the right-hand side to the background. In images generated by fingerprint sensors, the papillary lines are usually shown darkly while the background and the valleys between the papillary lines are shown in light colors. The histogram at hand is almost an ideal case since it is normally hard



to differentiate between the two hills. Very often, a hill is missing or the gray colors are distributed in such a way that there is only one continuous high curve to be seen in the histogram. The optimum would be for the mean value M to be in the center of the histogram, i.e. at the gray color 128 (when looking at 255 gray colors). In this ideal case the area below the left part of the curve SL equals the area below the right part of the curve SR.. Thus, the average value can be considered as a parameter for the quality of separation between the papillary lines and the background. The average lies in the area [0.256] since it represents a gray color. This situation is shown in Figure 4.5.3. To test how well this condition is fulfilled, the areas on the left-hand side and on the right-hand side have to be calculated. These areas are called SL and SR. They are calculated as follows:

∑=−

=

1

0

M

inL ihS )(

and ∑==

255

MinR ihS )(

M is our mean value (starting with M = 128) and hn(i) is the normalized frequency for the respective gray colors. If SL < SR, then the mean value will be incremented, i.e. it is moved to the right. If SL > SR, the mean value is decremented. In case SL ≅ SR, then the calculation of M is completed.

uenc

y

SL

SR

M

Grayscales

Nor

mal

ized

freq

Figure 4.5.3: Search for the mean value

To calculate the mean value, the limits of the histogram have to be taken into consideration. The beginning of the histogram shall be called BDark and the end of the histogram is called BBright. Consequently, the mean value should lie in the middle between the beginning and the end. This theoretical mean value will be denoted MT and can be calculated according to the following formula:

2BrightDark

T

BBM

+=

By using the real mean value M, the deviation A can be calculated as a percentage.

%1001 ⋅−=MM

A T



Example: The histogram starts with the gray color BDark = 20 and ends with the gray color BBright = 200. The expected mean value should be at the gray color MT = 110. According to the calculation (confer previous page), the mean value is MR = 140. The deviation A can be calculated by using these two mean values. The result is A = 127%, which means that the exceeding is 27%, which is, e.g. above the tolerance (value ± 20%). In Figure 4.5.4 there are examples of different images of all sensors, which were used in the test, and the respective normalized histograms in which the mean values are highlighted. It can be seen clearly that previously binarized images (which do not have different grayscales but rather two colors only, i.e. black and white) do not show any deviations, for example in the case of Sensor C.

Sensor A A = 79.1%

Sensor B A = 13.27%

Sensor C A = 0%

Sensor D A = 37.3%

Sensor E A = 22.6%



Sensor F A = 39.5%

Sensor G A = 6.79%

Sensor H A = 5.26%

Sensor I A = 59.38%

Sensor J A = 77.69% Figure 4.5.4: Examples of images from all sensors and their histograms

4.5.3 Separability

This method does not require previous knowledge about image contents and the binarization threshold S is determined according to a statistic quality criterion, which is taken from the discriminant analysis. The gray value image, which needs to be binarized, is described by the 1st degree statistic, i.e. by its standardized histogram p(0), p(1), ... p(255), with 0 ≤ p(i) ≤ 1 with p(i) being the probable frequency of the gray value i. Every possible threshold S with 0 ≤ S ≤ 255 separates the gray value into two classes K0 and K1, that is classes of gray values which, upon binarization, are allocated to the object and the background respectively. The probability of occurrence of the elements of this class is:

∑=

=S

iipP

00)( and

0

255

111)( PipP

Si−== ∑

+=



In addition, the following quantities need to be calculated: the mean gray values m0 of K0, m1 of K1, and the mean value m of the entire image with

m = m0P0 + m1P1

and its respective variations

∑=

⋅−=S

iipmis

0

2

0

2

0)()( ∑

+=⋅−=

255

1

2

1

2

1)()(

Siipmis

These quantities are all functions of parameter S. The criterion for determining S is the maximization of variance s2

B between K0 and K1 as well as the minimization of variance s2W within K0 and K1, i.e.

the search for preferably compactly bundled and well separated gray value classes. For this purpose, the quantities

2

11

2

00

2 )()( mmPmmPsB

−+−= 2

11

2

00

2 sPsPsW

⋅+⋅=

are calculated and the ratio (sB2 / sW

2), which depends on the parameter, is being maximized. The desired threshold is the value of S which maximizes this ratio.

4.6 U5 + U6 – Influences on the Fingerprints

4.6.1 U5 – Influence of the Sensors on the Fingerprints

Of each fingerprint of a test person, 3 fingerprint images were made at 3 appointments. Hence, 9 fingerprint images of the same finger are available (all fingers with the exception of the small finger). Since the images were acquired at different appointments and at different times, there have to be differences. Generally, slightly different regions of the area of the fingers may have been recorded. Soiling of the sensor surface may vary as well as temperature, humidity, incidence of light, and other environmental conditions. These environmental influences affect the quality with which the sensors acquire the images and, consequently, the acquired fingerprint images. Hence, it is possible to draw conclusions about the quality of the sensors by evaluating the image quality, which is also closely related to the evaluation of the fingerprint image quality in examination U4. In this case, the fingerprint images are analyzed by an algorithm (ME / MSA) prior to being processed. One criterion for accepting a fingerprint for verification purposes is the quality of the image. This is further described in Chapter 4.5 (U4 – Influence of the Sensors on Fingerprint Image Quality). There, three methods for quality evaluations were suggested: contrast, mean value, and separability of a given gray area. Additionally, the ridge count (number of lines) in a fingerprint image can be tested. The ridge count gives hints with regard to the capability to differentiate between the elevations of the papillary ridges and their valleys. This capability is a quantitative factor, which is defined as a rough number of lines that either touch or cross the imaginary connection between core and delta (confer Figure 4.6.1).



Figure 4.6.1: Ridge count

Due to the high complexity of different line configurations, an exact determination of the number of lines is often very difficult to be done. In the investigation at hand, a delta within the fingerprint image is missing in most cases since core and delta are only available in rolled fingerprints (for dactyloscopic purposes). In order to utilize the ridge count for a quantitative evaluation anyway, its definitions can be varied by examining the number of lines between two defined minutiae. However, an automatic evaluation of the number of lines between two minutiae largely depends on the feature extraction (ME) algorithm and its adaptation to a specific sensor. In order to eliminate the dependence between sensor and algorithm, we stop here by simply stating whether valleys and ridges of papillary lines can still be differentiated. Therefore, the fingerprint images were evaluated by an analyst for every sensor with regard to the separation between valleys and ridges. Grades were given in the following categories • very good separation (7), • good separation (3), and • moderate separation (1).

The results of these examinations are summarized in Table 4.6.1.

Ridge count possible: (grade)DB1 Yes (1) DB2 Yes (3) DB3 Yes (7) DB4 Yes (3) DB5 Yes (1) DB6 Yes (1) DB7 Yes (1) DB8 Yes (1) DB9 Yes (3) DB10 Yes (1) DB11 Yes (1)

Table 4.6.1.: Possibility of ridge count



4.6.2 Influence of Feature Extraction on the Fingerprints

The characteristics (i.e. parameters) of feature extraction do influence the quality of the fingerprint images. Again (confer Section 4.3), an evaluation based on comparing the characteristics is, unfortunately, not possible since there is no standardization for fingerprints. Table 4.6.2 shows a summary of parameters used by the algorithms for feature extraction purposes.

Attribute \ Algorithm Algorithm A Algorithm B Algorithm C

DIN V 66400

Extracted minutiae

G, L G, L, C, D G, L G, L

Angle (step size)

yes (not available) yes (not available) yes (not available)

yes (1.40625°)

Minimum number of minutiae

3 10 – 12 63 not available

Maximum number of minutiae

not available not available 191 not available

Rotation ability optional (step 5°) optional ± 30° not available

Translation ability not available optional not available not available

Other aspects image region of the minutiae and papillary line

width

quality, curvature, ridge density

good zones and zones that

cannot be used

not available

Attribute \ Algorithm Algorithm D Algorithm E Algorithm F Algorithm G Extracted minutiae

G, L not available L G, L

Angle (step size)

yes (not available) not available yes (not available)

yes (1.40625°)

Minimum number of minutiae

not available not available 5 not available

Maximum number of minutiae

100 not available 64 1024

Rotation ability not available not available −{}−± 45° 180 ° Translation ability not available not available 75 % not available

Other aspects grayscale and region information of the

minutiae

not available not available curvature of the papillary line in the minutiae;

singular points; block orientations

Table 4.6.2.: Algorithm parameters

Meaning of the individual data: • Extracted minutiae: G = ridge bifurcation, L = ridge ending • Angle of the minutiae (gradient) and step size (0° to 360°) • Minimum number of minutiae: How many minutiae are required for successful verification. • Maximum number of minutiae: How many minutiae can be used by the MSA algorithm.



• Rotation ability: By how many degrees can the two fingerprint images or the corresponding number of minutiae be rotated towards each other? The percentages have to be considered in relation to the sensor dimensions.

• Translation ability: By how many percent can the two fingerprint images or the corresponding number of minutiae be shifted toward each other.

• Other aspects: What other things are extracted from the fingerprint image and then used for verification purposes.



5 Test Results of Various Systems

5.1 Introduction

The chapter at hand contains the test results . Individual characteristics of the systems (sensors) are described in chapter 4. These sections contain the name, classification (access system / dactyloscopic system), a picture of the sensor, dimensions of the sensor and device as well as additional information. A description of the significance of the ROC curves is found in Chapter 2 of this report. The result of comparing two fingerprint images is a similarity score, the so-called matching score. In order to generate a graph with curves for both authorized and unauthorized persons all fingerprint images in the database have to be compared with each other. The scores of matching fingerprint images of the same finger generate the probability density for authorized persons. Fingerprint images of different fingers (persons) generate the probability density for unauthorized persons as a function of the score. Depending on the threshold that was set, there are different error rates. Ideally, the densities do not overlap and a threshold T can be set where no errors occur. Based on the inadequacy of existing systems, errors will occur both with regard to false acceptance FAR(T) and with regard to false rejection FRR(T). T can lie anywhere between 0 and 1. The presentation of FRR(T), being a function of FAR(T), eliminates threshold T as a parameter and thus makes an objective comparison of various systems possible, even if they have completely different probability densities. This so-called ROC curve is the basis for comparing different fingerprint systems such as required in BioFinger. These curves are added at the end of each section. Altogether, there is a matrix of all possible ROC curves for all sensors and algorithms - Confer Table 5.1.1.

Figure 5.1.1: Matrix of the ROC curves

One line of the matrix contains all ROC curves for a fixed algorithm and all sensors if they are supported by the algorithm. The columns contain ROC curves for a fixed sensor and different algorithms. In order to generate the curves, all images of all sensors were evaluated by using all algorithms. These processing results are stored in this matrix as a curve. Alongside one column of the matrix, all ROC curves can be inserted into a graph. Thus, the question can be answered with which algorithm the sensor is best compatible, i.e. the best ROC curve in a graph with several ROC curves for different algorithms can be chosen. (Number 2 in Figure 5.1.1). The same procedure can be carried out for algorithms. Alongside a row of the matrix, the ROC curve can be inserted into a graph. Hence, the question can be answered with which sensor the algorithm works best – i.e. the best ROC curve in this graph with several ROC curves for different sensors can be chosen. (Number 1 in Figure 5.1.1).



5.2 Evaluation of the Fingerprint Quality

In order to evaluate the fingerprint image quality, different metrics such as contrast, percentage of mean value deviation in the histogram as well as separation ability were examined. The fingerprint image quality is, however, not defined by the image quality but in terms of the verification quality, i.e. by performance or measured error rates. Since the verification performance is reflected in the ROC curves, the latter have to be evaluated with regard to the algorithm that was used. For this reason, the chronology of the ROC curves was arranged starting with the worst and ending with the best performance for each of the respective algorithms. Afterwards, different metrics were inserted as a function of this ranking of the various sensors per fixed algorithm (confer Figure 5.2.1).

1 2 3 4 5 6 7 8 9 10 0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

1.1

Ranking

Metrics

Figure 5.2.1: Quality metrics as a function of ranking

If there is a correlation between the respective quality and the fingerprint image generated by the sensor, this should be visible in these graphs depending on the respective algorithm. The following graphs show different quality metrics as a function of the ranking according to the ROC diagram, using algorithm No. 1 as an example. The remaining algorithms show the same results in terms of the correlation between the used metrics and the verification performance. If there is a correlation between quality metrics and verification, a higher ranking should indicate a higher quality of the fingerprint image and thus of the sensor. However, this conclusion could not be drawn for any of the various algorithms with the metrics that were used.



0 2 4 6 8 10 120.4

0.5

0.6

0.7

0.8

0.9

1Algorithm 1

Ranking

Contrast

0 2 4 6 8 10 12-20

0

20

40

60

80

100

120

Ranking

Percentage deviation

Algorithm 1

0 2 4 6 8 10 120.6

0.7

0.8

0.9

1

1.1

1.2

1.3

1.4

Ranking

Separability Algorithm 1



Figure 5.2.2: Quality as a function of ranking (contrast, percentage deviation, separation ability) for algorithm No. 1

5.3 Comparison of the Systems

In order to compare different systems, a ROC curve can be chosen for each sensor; then it can be inserted into a plot. This results in the combination of various systems sorted according to their performance, i.e. according to the lowest error rates. This is shown by a path going across the matrix of the ROC curve (confer Figure 5.3.1).

Sensor 1 ... Sensor i ... Sensor M Algorithm 1 ROC11 ... ROCi1 ... ROCM1 ... ... ... ... ... ... Algorithm j ROC1j ... ROCij ... ROCMj ... ... ... ... ... Algorithm N ROC1N ... ROCiN ... ROCMN

Figure 5.3.1: Path in the matrix

In Figure 5.3.2, various systems are compared.

10-3

10-2

10-1

10010

-3

10-2

10-1

100

FAR

FRR

S1 A7EERS2 A4S3 A6S4 A7S5 A7S6 A7S7 A6S8 A6S9 A4S10 A7S11 A7S13 A6

S4

S1

S3

S5 S6

S7

S8

S9 S10

S2 S11 S13

Figure 5.3.2: Best ROC curves for all sensors

The choice of the best ROC curve for any given sensor is not always easily made since the best curve is not necessarily located nearest to zero for all value pairs (FRR, FAR).



Algorithm Sensors ALG7 1, 4, 5, 6, 10, 11 ALG6 3, 7, 8, 13 ALG4 2, 9

Table 5.3.1: Algorithm-scanner combinations

Furthermore, depending on the envisaged application, we are moving along fixed working lines (same errors, ERR – Confer Fig. 5.3.2) or in certain areas of the ROC curve (confer Section 3.3). Therefore, the evaluation below was carried out for the working lines FAR = FRR (EER) and a fixed FAR = 0.01 as well. The results for various sensors and algorithms are documented in the tables below.

DB1 DB2 DB3 DB4 DB5 DB6 DB7 DB8 DB9 DB10 DB11 DB13 ALG1 0.028 0.174 0.225 0.049 0.044 0.046 0.128 0.025 0.097 0.050 0.165 0.123 ALG2 0.013 0.112 0.171 0.026 0.022 0.038 - 0.095 - 0.027 0.074 - ALG3 0.049 0.283 0.268 0.074 0.118 0.185 0.056 0.042 0.440 0.072 0.106 - ALG4 0.003 0.102 0.175 0.013 0.023 0.021 0.011 0.006 0.027 0.023 0.054 0.120 ALG5 0.023 0.304 0.240 0.107 0.042 0.082 - 0.021 - 0.102 0.119 0.107 ALG6 0.008 0.269 0.140 0.035 0.024 0.028 0.016 0.009 0.032 0.032 0.065 0.049 ALG7 0.004 0.254 0.387 0.017 0.014 0.025 0.049 0.044 0.029 0.021 0.049 0.075

Table 5.3.2: EER for sensor – algorithm combinations (1 = 100 %)

DB1 DB2 DB3 DB4 DB5 DB6 DB7 DB8 DB9 DB10 DB11 DB13 ALG1 0.034 0.287 0.361 0.065 0.065 0.061 0.182 0.030 0.143 0.066 0.248 0.232 ALG2 0.014 0.182 0.319 0.034 0.026 0.049 - 0.095 - 0.034 0.098 - ALG3 0.081 0.459 0.492 0.114 0.157 0.274 0.074 0.055 0.702 0.132 0.164 - ALG4 0.006 0.145 0.220 0.023 0.022 0.024 0.020 0.011 0.030 0.029 0.078 0.144 ALG5 0.024 0.449 0.315 0.123 0.052 0.089 - 0.022 - 0.113 0.133 0.156 ALG6 0.008 0.487 0.230 0.047 0.029 0.034 0.018 0.009 0.042 0.039 0.082 0.064 ALG7 0.004 0.631 0.783 0.020 0.015 0.028 0.082 0.068 0.039 0.023 0.089 0.128

Table 5.3.3: FRR(FAR = 0.01) for sensor – algorithm combinations (1 = 100 %)

In connection with this please note that not all algorithms for all sensors could be tested since the algorithm manufacturer could not adapt the fingerprint scanner on time. This can be seen in the missing pillars for certain algorithm-sensor combinations in the graph on the following page and the empty cells in the tables above.



Figure 5.3.3: EER for algorithms and sensors



Figure 5.3.4: FRR with FAR = 0.01 for all algorithms and sensors



5.4 ROC Curve for Sensor 1

10-3

10-2

10-1

100

10-3

10-2

10-1

100 SENSOR 1

FAR

FRR

A1A2A3A4A5A6A7




10-3

10-2

10-1

100

10-3

10-2

10-1

100

FAR

FRR

SENSOR 2

A1A2A3A4A5A6A7




10-3

10-2

10-1

100

10-3

10-2

10-1

100

FAR

FRR

SENSOR 3

A1A2A3A4A5A6A7




10-3

10-2

10-1

100

10-3

10-2

10-1

100 SENSOR 4

FAR

FRR

A1A2A3A4A5A6A7




10-3

10-2

10-1

100

10-3

10-2

10-1

100 SENSOR 5

FAR

FRR

A1A2A3A4A5A6A7


10-3

10-2

10-1

100

10-3

10-2

10-1

100 SENSOR 6

FAR

FRR

A1A2A3A4A5A6A7




10-3

10-2

10-1

100

10-3

10-2

10-1

100 SENSOR 7

FAR

FRR

A1A3A4A6A7


10-3

10-2

10-1

100

10-3

10-2

10-1

100

SENSOR 8

FAR

FRR

A1A2A3A4A5A6A7




10-3

10-2

10-1

100

10-3

10-2

10-1

100

SENSOR 9

FAR

FRR

A1A3A4A6A7


10-3

10-2

10-1

100

10-3

10-2

10-1

100

SENSOR 10

FAR

FRR

A1A2A3A4A5A6A7




10-3

10-2

10-1

100

10-3

10-2

10-1

100

SENSOR 11

FAR

FRR

A1A2A3A4A5A6A7


5.15.1 Description of the System

• Description: fingerprint images of the BKA database • Figure: example image

• Resolution [dpi]: 500 • Image size: 780 × 780



5.15.2 ROC curve

10-3

10-2

10-1

100

10-3

10-2

10-1

100

SENSOR 13

FAR

FRR

A1A4A5A6A7

Figure 5.15.1: ROC curves for Sensor 13



5.16 ROC Curve for Algorithm 1

10-3 10-2 10-1 10010-3

10-2

10-1

100ALGORITHM 1

FAR

FRR S1S2S3S4S5S6S7S8S9S10S11S13




10-3 10-2 10-1 10010-3

10-2

10-1

100ALGORITHM 2

FAR

FRR

S1 S2 S3 S4 S5 S6 S10 S11 S8




10 -3 10-2 10-1 10010-3

10-2

10-1

100 ALGORITHM 3

FAR

FRR

S1 S2 S3 S4 S5 S6 S7 S8 S9 S10 S11




10-3 10-2 10-1 10 010-3

10-2

10-1

100 ALGORITHM 4

FAR

FRR

S1 S2 S3 S4 S5 S6 S7 S8 S9 S10 S11 S13

S9

S5S4 S6 S10 S7




10-3

10-2

10-1

10 010

-3

10-2

10-1

100 ALGORITHM 5

FAR

FRR

S1 S2 S3 S4 S5 S6 S8 S10 S11 S13




10 -3 10-2 10-1 10010-3

10-2

10-1

100

FAR

FRR

ALGORITHM 6

S1S2S3S4S5S6S7S8S9S10S11S13




10-3

10-2

10-1

100

10-3

10-2

10-1

100

FAR

FRR

ALGORITHM 7

S1 S2 S3 S4 S5 S6 S7 S8 S9 S10 EERS11 S13



6 Investigations with the Fingerprint Database

6.1 Description of the Databases

In order to examine the differentiability of fingerprint images and the influence of ageing effects of fingerprints on verification, two data records were specially selected by the Federal Office of Criminal Investigation (BKA). One data record consists of seven lists with a varying number of fingerprints of different persons which, when compared with the AFIS system of the BKA, showed significant similarities. The second data record comprises fingerprints of a total of 183 persons gathered over an extended period of time. The number of acquisitions as a function of the respective year can be seen in the graph (confer Fig. 6.1.1).

1940 1950 1960 1970 1980 1990 2000 2010

0

10

20

30

40

Frequency distribution

Num

ber o

f acq

uisi

tions

Year of acquisition

Figure 6.1.1: Total number of acquired fingerprints depending on the year of acquisition

The available time intervals between forensic processing operations can be seen in Figure 6.1.2 in the values on the x-axis.



0 10 20 30 40 50 60 0 20 40 60 80

100 120 140 160 180

Time interval

Number

Figure 6.1.2: Number of acquisitions per time interval

Furthermore, the number of acquisitions for the respective time period can also be seen. If the number decreases depending on the time difference between forensic processing, the result is that less fingerprint images for matching identical fingers are available. Consequently, if the ages of fingerprints differ 30 years, then the statistic significance of a comparison will be much lower than if the age difference is only 10 years. In addition, fingerprint images that are 10, 20, or 30 years old do not exist for every person.



0 10 20 30 40 50 600

10

20

30

40

50

60

70

80

90

100

Time interval

Number of persons

Figure 6.1.3: Number of persons with respective time intervals

For this test, only 3 algorithms were used since only 3 manufacturers adapted the configuration files for this test. According to Figure 6.1.3, there are fingerprint images taken at intervals of 10, 20, or 30 years for the number of persons indicated in the table below:

Time interval Number of persons 10 65 20 55 30 26

6.2 Research on the Differentiability with Similar Fingerprints

6.2.1 Description of the Examination

Biometric recognition minutiae are both unique and naturally unchangeable (a basic fact of dactyloscopy). The fact that different fingerprint images can be classified and differentiated using biometric systems is a significant prerequisite for utilizing this biometric characteristic. In order to test the performance of different algorithms with regard to differentiability, special fingerprints were selected from BKA's database. These fingerprints are from different fingers even though they show similarities that make it difficult for biometric systems to differentiate between these images (so-called



"biometric twins"). Therefore, the biometric system can be misled by similar fingerprints and thus wrongly conclude that fingerprints of different fingers come from the same finger. Since the data record was selected with the AFIS system of BKA, the measuring results that were achieved have to be seen in relation to the performance of the AFIS system. Since all of these images were rolled, the area of these fingerprint images is naturally larger than that of fingerprints that were made by laying on fingers. In the case of rolled fingerprint images, both of the principal fingerprint features, i.e. delta and core, are to be found. Some images show very similar minutiae patterns. The greatest similarity is found in the position of the minutiae. If the algorithm does not use additional data such as fingerprint type, angle, or pattern, different fingerprint images with similar positions can be evaluated as coming from the same finger. When using these additional data, it is less likely that the fingerprints are placed in the wrong category. Contrary to access control systems, dactyloscopic systems carry out an additional pattern classification of the fingerprints (categorization of the fingerprint into a pattern class based on the raw papillary line pattern in the entire fingerprint; pattern classes are: arch, loop, and whorl). Most access control systems do not feature pattern classification. Rather, they simply extract minutiae and differentiate only two kinds of minutiae (ridge ending and ridge bifurcation).

Figure 6.2.1: Example of similar fingerprints

In order to test the performance of various algorithms with regard to their differentiability, different fingerprint images selected by the BKA were used as a database. For the purpose of comparing different fingerprints (index u) of biometric twins (index z), the probability density pu

z(s | Hu) of the similarity values was determined in an experiment. In order to determine the probability density pg(s | Hg) of the similarity values of identical fingerprints (index g), fingerprint images of the same persons were chosen from the database where the interval between taking the fingerprints was rather short. This was done in order to avoid that the examination of the differentiability would be mixed with the examination of the ageing properties when matching identical fingerprint images (confer section below).

For the purpose of examining differentiability, the fingerprints of different persons, which do not show great similarities, were matched in order to generate a second probability density pu(s | Hu) which can be used to compare it with the respective algorithm (confer Figure 6.2.2).



( )uu Hsp | ( )uz

u Hsp |( )gg Hsp |

0

)( Hsp

S

Figure 6.2.2: Dependence of probability densities when comparing biometric twins

The probability density pg(s | Hg) of the score values of identical fingerprint images (index g) does not change. Therefore, if the function pu(s | Hu) ⇒ pu

z(s | Hu) changes when comparing biometric twins, a shift in the ROC curve should occur along the FAR-axis (Confer Figure 6.2.3).

FRR

FAR

1

1

Δt1

Δt2

Figure 6.2.3: Shift in the ROC curve when comparing biometric twins

Thus, changes in the probability density for comparing different fingers with the respective algorithm can be determined. In addition, the change in the ROC curve and thus the error rates for each and every algorithm can be examined as well. Furthermore, different algorithms can be compared with each other with regard to their capability of differentiating between similar fingerprint images if all ROC curves are entered into a graph.

6.2.2 Results

The measurements for the individual algorithms were carried out according to the description above. As described in Chapter 2, the algorithms that were used were named ALGi , i = 1,...,NALG. ALGi. Below, the results of the tests are shown using the algorithms indicated above. Unfortunately, two algorithms are missing in this test since they could not be adapted to the BKA database prior to the beginning of the test. The hypothesis of the test according to the description above was confirmed for nearly all algorithms. When comparing similar fingerprint images that came from different fingers the performance of all algorithms was reduced apart from Algorithm 7. The performance of Algorithm 6,



however, is still better, when comparing biometric twins, than the performance of the remaining algorithms (confer Figure 6.2.9).

10-3 10-2 10-1 10010-3

10-2

10-1

100

FAR

FRR

Original dataBiometric twinsEER

Figure 6.2.4: Shift in the ROC curve when comparing biometric twins for Algorithm 1

10-3

10-2

10-1

100

10 -3

10 -2

10 -1

10 0

FAR

FRR

Original data Biometric twins ERR




10 -3 10-2

10-1

100

10-3

10-2

10-1

100

FAR

FRR

Original data Biometric twins EER


10 -3 10-2

10-1

10 010

-3

10-2

10-1

100

FAR

FRR

Original dataBiometric twinsERR




10-3

10-2

10-1

100

10-3

10-2

10-1

100

FAR

FRR

Original data Biometric twins EER




10-3

10-2

10-1

10 010 -3

10 -2

10 -1

10 0

FAR

FRR

Algorithm 1Algorithm 4Algorithm 5Algorithm 6Algorithm 7ERR

Figure 6.2.9: Comparison in the performance of the algorithms when comparing biometric

twins

6.3 Research on the Ageing Characteristics of Fingerprints

6.3.1 Description of the Examination

The investigation described in this section focuses on fingerprint images coming from the same finger. In addition, it is only the ageing effect of the fingerprints that is to be examined. In most cases, the ageing process does not change the structure of the fingerprint image. The papillary lines always show the same pattern since the information thereof is stored in the epidermis. If a wound was sustained that was not very deep, i.e. where only the upper skin was injured, after a certain time, the same papillary lines are formed as before. Even the ageing process cannot change the paths of the papillary lines. The fingerprint may be a little larger, the papillary lines may be lower (if they were worn due to working), and the finger may show some wounds. However, the structure always remains the same. Therefore, it should not be difficult, for the different verification algorithms, to identify fingerprints of the same finger, which only differ in the date of their acquisition.



Figure 6. 3.1: Example of a fingerprint of the same finger acquired at different dates (top left

to bottom right: 1960, 1962, 1980, 2001)

In order to test the influence of ageing, fingerprint images of the same finger were selected from the database in order to determine the probability density pg

Δt(s | Hg) of the score values s for the purpose of comparing fingerprint images of the same finger (index g) as a function of the time interval (index

). For determining the probability density tΔpu(s | Hu) of the score values of different fingerprint images (index u), the year in which most test persons were enrolled was chosen from the data record in order to keep the time intervals between the acquisitions of different persons as small as possible (Confer Figure 6.3.2).



( )g tg H sp |1Δ

( )gt

g Hsp |2Δ( )uu Hsp |

0

)( H sp

S

years tyears t

12 852

2

1

−=Δ−=Δ

Figure 6. 3.2: Time-related dependence of the probability densities pgΔt(s | Hg) when

comparing identical fingers

The probability density pu(s | Hu) of the score values of different fingers (index u) does not change. Therefore, if the function pg(s | Hg) ⇒ pg

Δt(s | Hg) is changed when comparing fingerprints of the same finger taken at different points in time, a shift in the ROC curve along the FRR-axis should occur (Confer Figure 6.3.3).

FRR

FAR

1

1

Δt1

Δt2

Figure 6. 3.3: Shift in the ROC curve due to the effects of ageing

Thus, changes in the probability densities pgΔt(s | Hg) for comparing identical fingers can be

determined for different algorithms as a function of the time interval at which the fingerprints were taken. Furthermore, the change in the ROC curves and thus the error rates can be examined for every algorithm. In addition, different algorithms can be compared with regard to their robustness against ageing of identical fingerprint images if all ROC curves are plotted in one graph.

6.3.2 Results

The results of the individual algorithms were carried out according to the description above. As described in Chapter 2, the algorithms that were used were named ALGi , i = 1,...,NALG. The results shown below are obtained by using the algorithms indicated above. Unfortunately, two algorithms are missing in this test since they could not be adapted to the BKA database prior to the



beginning of the test. The tests confirmed the general hypothesis of the section above for all algorithms. When comparing similar fingerprint images that came from the same fingers, but were acquired at different times, the performance of all algorithms was reduced.

10 -2

10-1

10 0

10 -2

10 -1

10 0

FAR

FRR

1-5 years8-12 years18-22 years28-32 years

Figure 6. 3.4: Shift in the ROC curve due to the ageing effect with regard to Algorithm 1



10 -2

10-1

10 0

10 -2

10 -1

10 0

FAR

FRR

1-5 years 8-12 years 18-22 years 28-32 years




10-2

10-1

100

10-2

10-1

100

FAR

FRR

1-5 years 8-12 years 18-22 years 28-32 years




10 -2

10 -1

10 0

10 -2

10 -1

10 0

FRR

1 -5 years8 - 12 years 18 - 22 years 28 - 32 years EER

FAR Figure 6. 3.7: Shift in the ROC curve due to ageing effect with regard to Algorithm 6



10-2

10-1

100

10-2

10-1

100

FAR

FRR

1 - 5 years 8 - 12 years 18 - 22 years 28 - 32 years EER


In order to measure the relative increase in error rates, a quotient of the FRR values for different time intervals was computed as a function of the FAR (for identical FAR values). The quotient hence shows the relative change in the FRR when comparing two ROC curves for different time intervals.



0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 1.2

1.4

1.6

1.8

2

2.2

2.4

2.6

2.8

FAR

q

10 years 30 years

Figure 6.3.9: Relative change of the FRR with regard to Algorithm 1

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 1.5

2

2.5

3

3.5

4

4.5

5

5.5

6

6.5

q

10 years 30 years




0 0.2 0.4 0.6 0.8 10

0.5

1

1.5

2

2.5

3

FAR

q

10 years 30 years


0 0.2 0.4 0.6 0.8 1 0.5

1

1.5

2

2.5

3

3.5

4

4.5

FAR

q

10 years 30 years




0 0.2 0.4 0.6 0.8 1 1

1.2

1.4

1.6

1.8

2

2.2

2.4

2.6

FAR

q

10 years 30 years


In conclusion, when comparing fingerprint images with a time difference interval ∆t2 = 10 years, it can be estimated that the FRR decreaes by factor 2q ≈ upwards.

6.3.3 Examination of Ageing According to Age Groups

According to the results of the above section, the question is whether the measured effects that occur, when examining ageing and the related degradation of the FRR, equally arise in all age groups, or whether the wear and tear of the ridges mainly concerns older people in the database. For this purpose, three age groups (30, 40, and 50 years) were formed. For each person within an age group fingerprints taken at time intervals of 10, 20, and 30 years were compared with each other.



10 -2

10-1

10 0

10 -2

10 -1

10 0 Age group 30, Algorithm 4

FAR

FRR

1-5 Years8-12 Years 18-22 Years28-32 Years

Figure 6.3.14: Shift in the ROC curve in the age group of the 30-year-olds for Algorithm 4



10 -2

10-1

10 0

10 -2

10 -1

10 0 Age group 40, Algorithm 4

FAR

FRR

1-5 Years8-12 Years 18-22 Years

28-32 Years

Figure 6.3.15: Shift in the ROC curve in the age group of the 40-year-olds for Algorithm 4

According to the curves in the above graph, a mixed picture can be drawn. The results do not indicate clearly whether any age group shows a significant degradation of the values compared to other age groups. The low number of fingerprints of the same person, which were taken at time intervals of 10, 20, and 30 years (confer Figure 6.1.3) is the reason for the limited number of comparisons in these examinations. The additional narrowing down to certain age groups further reduces the number of fingerprints included in the evaluation. Therefore, the results presented here are not suitable to draw reliable conclusions whether the ageing process with regard to a certain age group might have a greater influence on the similarity scores of individual comparisons. Since the results of the other algorithms are similar, only one algorithm was selected here as an example in order to document the results. In connection with this, a larger database and further examinations with regard to age groups are necessary in order to come to reliable results.

6.4 Simulation of the Reduced Sensor Area

As a question of costs, a mass application of fingerprint recognition systems in identification and verification scenarios may require the use of a large number of reasonably priced fingerprint sensors. Generally, the use of cheap sensors can result in a reduced usable sensor area for acquiring the fingerprint. In connection with this, the question is what kind of performance a biometric recognition system has, if, due to the reduced sensor surface, only a part of the fingerprint is available for comparison purposes. A reduction in the fingerprint that is available implicitly leads to the extraction of a lower number of features, which would result in an increase of the FNMR.



On the other hand, it is not to be expected that the FMR for comparing different fingers would improve when using a reduced usable sensor area, i.e. the probability density pu(s | Hu) of the similarity values of different fingerprint images (index u) ought not to change. One would expect, that if a negative effect of the error rates occurs, it should result in a change of the function when comparing fingerprint images of the same finger with a reduced usable sensor area (Confer Figure 6.4.1).

p

s

BerechtigtePerson

Nichtberechtigte

Person

BerechtigtePerson für

kleinere Fläche

10

Nicht berechtigtePerson für

kleinere Fläche

Different features

Different features for a smaller area

Same feature for a smaller area

Same feature

Figure 6.4.1: Change of the probability densities due to reduction of the usable sensor area

In general an increase in both error rates could happen. In that case the distribution of the score values by comparing identical fingerprints would shift to the left and the distribution corresponding to the score values of different fingerprints would shift to the right. Consequently, a reduction in the sensor area should be seen in an increase of the FRR and FAR (Confer Figure 6.4.2).

FRR

FAR

1

1

Δt1

Δt2

Figure 6.4.2: Change of the probability densities due to a reduction of the usable sensor area

For this examination, fingerprint images of BKA's database were used which contained a total of 183 persons. In order to simulate a reduced sensor area, a sector of 416 x 416 pixels was extracted from the original image, which was then compared with the template that was calculated from the original data (780 x 780 pixels). By using the original data, it was guaranteed that the effect of the reduced sensor



area would not be mixed with different acquisition conditions which would be the case in a field test if sensors with a smaller usable sensor area were to be used. Altogether it can be stated that cutting a section of the existing image data to the size of image data provided by the sensors that are available on the market does not lead to a degradation of the error rates. This result is consistent for all algorithms. If the matching fingerprint that is stored in the database was not rolled but is only a section of the fingerprint, no conclusion can be drawn by using these experimental results. In addition to this, further examinations are required, which should evaluate the degree of dependence on the size of matching fingerprint images stored in the database and the sensor area that was used.

10-2

10-1

100

10-2

10-1

100

Sensor 13Sensor 14

Figure 6.4.3: Simulation of a smaller sensor area with Algorithm 1



10-2

10-1

100

10-2

10-1

100

Sensor 13Sensor 14




7 Standards and Universal Fingerprints

Different systems use different kind of information found in fingerprints. With regard to their application, a rough differentiation can be made between dactyloscopic systems (AFIS) and access control systems. The systems in these two groups do not only use different sensors (by rolling fingers or by simply placing them thereon) but also different minutiae and different numbers of minutiae of the fingerprint. Furthermore, it has to be noted that there are great differences between various access control systems. Some systems only extract minutiae, whereas others use the regions around the minutiae (patterns) as well or they exclusively use image data. For this reason, it is impossible to transfer the stored templates of different systems into a uniform format. If this was possible, the match algorithms would have to be adapted accordingly. The argument of the companies, which is not in favor of changing their own templates, is the fact that the structure of and the operation mode in which the algorithms work would have to be changed. Although the smallest common denominator at which all algorithms could work does provide for an inter-operability of the different systems, it does so at the expense of the performance (i.e. it would result in higher error rates). The inter-operability of the systems will only be guaranteed if the fingerprint image is used.

7.1 Feasibility and Algorithm Methods

7.1.1 Feasibility

As it can be seen in the description of the examination in Section 4.3, the definition of a universal fingerprint image or template is currently not possible. The systems, which were tested in Section 4.3, store different data of the template (fingerprint image). Some applications only work with minutiae, others only with patterns, and others use a combination of the two. Algorithm 2, for example, only uses data about the minutiae. Algorithms 4 and 5 use data about the region surrounding the minutiae in addition. Currently, there are three basic directions (methods) to be seen on the market with regard to the development of fingerprint recognition systems. The first method is solely based on minutiae. These systems extract only minutiae from the fingerprint image in order to compare them with other fingerprint images. Hence, for these systems, it is only the minutiae that are stored in the templates. The second method is based on a pattern description (e.g. Gabor Filter). This method does not use minutiae but rather processes parts of the fingerprint image using a mask and then stores the respective image data. This data can neither be stored as vectors nor in image sections. Either a "normal" comparison is carried out or some kind of correlation for individual image sections found between these data volumes. The third method is a combination of the two previous ones. Minutiae, for instance, are saved and, in addition, the environment of the minutiae, i.e. an image section, is also stored. Since almost all other systems use other data for comparison purposes, different data necessary for matching purposes are stored. Furthermore, the same data are stored differently by different systems if these systems use grids of different sizes. These applications can use large (rough) grids, i.e. the position of the minutiae, for example, can be scaled in steps of 10-1mm. In other algorithms, the intervals are set at 10-2mm. The same applies to angles (gradients) (intervals in degrees). For individual systems, a standardization of all existing systems would imply the loss of a part of their data or data splitting up into smaller grids. Principally, it is possible to define a universal fingerprint image (template) (confer Chapter 7.2). However, the matching algorithms need additional data of the fingerprint in their own template format for matching purposes. If only the minutiae of a system, which runs by using additional data about the environment of the patterns, are being used, then the comparison can either not be made or the system performance is drastically reduced.



A universal fingerprint image can only be defined, if all systems refer to the same data for matching purposes. As an alternative, several universal fingerprint images could be defined for each group of algorithms. However, if it was only the methods presented at the beginning of this section that were used, we would at least have three different definitions of the universal fingerprint image, which would contradict the universality aim.

7.1.2 Algorithm Procedures

Today's access control systems were developed from dactyloscopic systems. The method of these systems was similar to the experts' analysis of fingerprint images. First, the fingerprint was classified and then some minutiae therein were detected, which were also drawn on the fingerprint image. Then the expert could match up two fingerprint images, first the finger class and then the detected minutiae. Later on, computer technology was also used in order to make this hard work easier. In these systems, the minutiae were stored in templates. These templates contain the finger class of the fingerprint image as well as the detected (extracted) minutiae. Naturally, further additional data (for forensic purposes), such as e.g. name, age, site of the crime, ID (card) number, etc. were stored. These working stations had (and still have) a very high performance and were not limited to the capacity of the storage space. Proposals were made to use this system (with some changes) for access control systems. It was clear that these systems already showed a lower performance and it became necessary to reduce the volume of the templates. Initially, the data was only stored on the hard-drive (or on discs), i.e. there was always enough memory space. With the introduction of the smart card as a data carrier the capacity of the memory was limited. Therefore, the number of characteristics had to be reduced for storing purposes. At first, the number of minutiae was filtered, i.e. the minutiae at the fringe as well as imprecise minutiae were deleted. In addition, the data had to be compressed. Nowadays, it is not only the minutiae that are extracted from the fingerprint image but also additional data, which contain e.g. sections of the image in their templates. Today, different systems are available on the market, which use different fingerprint recognition methods. These systems involve a large number of templates. Compatibility is not guaranteed. Therefore, it is currently not possible to suggest or define a universal fingerprint standard (template) which can be stored on a smart card.

7.2 Documentation of Standards

The summary at hand consists of two parts. In the first part, general standards are presented which are, among other things, relevant to fingerprint recognition. In the second part, special standards for fingerprint recognition technologies are described.

7.2.1 General Standards

The BioAPI Specification

Title: JTC 1/SC 37, CD 19784, Information Technology, The BioAPI Specification This specification defines the Application Programming Interface (API) and the interface for Biometric Service Providers (BSP). The specification does not examine safety requirements for biometric applications or for providers even though these aspects are addressed at some points of the specification. The purpose of BioAPI is to make a general biometric authentication model available. This authentication model shall be suitable for all biometric technologies. The model covers such basic functionalities as enrollment (enrolling biometric minutiae into the database), verification, and identification. Furthermore, the model contains a database interface, which enables providers of



biometric services to manage all functionality needed for identification purposes in such a way that the best possible performance can be achieved. In addition, the model contains some elementary functions, which make it possible for an application to acquire the minutiae on the Client and to carry out enrollment, verification, and identification on the server. BioAPI can provide for the following: • Simple application interfaces • Standardized modular access to biometric functions, algorithms, and devices • Safe and robust management and storing of biometric data • Standard methods for differentiating between biometric data and device types • Supporting biometric identification in distributed systems BioAPI is an interface derived from individual biometric technologies, concrete manufacturer implementations, certain products, and devices. The description is made on a very high abstraction level, in order to guarantee that it is suitable for many potential software systems. Different standard interfaces for access to biometric mechanisms are described. BioAPI was developed in such a way that it is suitable both for application developers as well as for developers of biometric technologies. Among other things, the specification deals with the following aspects: BioAPI data structures, BioAPI registration scheme, BioAPI error handling, BSP operations (BSP = Biometric Service Provider), BioAPI interface for service providers, data structures for service providers, and operations of a service provider. Pursuant to ISO/SC37, the BioAPI specification is dealt with in WG 2 (Working Group on Biometric Technical Interfaces) as CD 19784. At the second plenary meeting of SC37 in Rome, the BioAPI service provider was awarded the Final Committee Draft (FCD) status. Common Biometric Exchange Framework Format (CBEFF)

Title: JTC 1/SC 37, CD 19785.2, Information Technology, Common Biometric Exchange Framework Format Originally, CBEFF stood for Common Biometric Exchange File Format. The Common Biometric Exchange Format Framework (CBEFF) describes data elements, which will be required for generally supporting biometric technologies. This data can be stored in a single file, which is used for exchanging biometric information between different system components or between systems. Since an exchange of biometric data becomes possible, CBEFF supports the inter-operability of biometrics-based application programs and systems, which were developed by different manufacturers. The specification addressed here is a revised version of the original version of CBEFF, which was published under the name NISTIR 6529. CBEFF offers an upward compatibility for technological improvements and provides for establishing new formats. It simplifies the integration of software and hardware, which was produced by different manufacturers. Furthermore, CBEFF describes data fields which are "necessary" or "optional" as well as the “sector of application” in order to indicate in which cases standards can be used or where a CBEFF-compatible specification is available. CBEFF can be used for every biometric technology and contains the description of formats and the content for these data elements such as: • Header, which contains data about the version number, length of data, and whether data is

encoded • Biometric data • Biometric basic data structures (head data + biometric data) Furthermore, CBEFF describes nested structures of biometric data. he CBEFF specification contains a description of the CBEFF structure, in order to reach the goal of clearly identifying the format and the author of all data structures within a CBEFF-compatible



structure. For this purpose, standard data structures are described (head data and possible data elements) as well as nested data structures. One section of the specification deals with the topic "CBEFF Patrons and Clients”. The CBEFF "Patron" is an organization that has defined a standard or a specification with regard to biometric data objects, which comply with CBEFF requirements. These are organizations such as BioAPI Consortium, ANSI Subcommittee X9, Working Group F4, etc. The CBEFF client is a unity, which has developed a specific biometric data structure in compliance with CBEFF requirements. The clients can be suppliers, working groups, or an industrial consortium. In ISO/SC37, the CBEFF specification is being processed in WG 2 (Working Group on Biometric Technical Interfaces) as the 3rd CD 19785.2. Presently, there are still a number of comments and proposed changes. Developing Biometric Profiles (White Paper)

March 12, 2003 Title: US Contribution to SC 37/SG 4 on Developing Biometric Profiles This publication proposes a method for developing biometric application profiles. The document deals with the implementation of a prepared analysis in order to identify universal functions which, together, support the entire series of biometric scenarios. Afterwards, these functions and their interdependencies are mapped on generic application classes and on specific elements of the basic standards. This publication is a framework which can be used to define functions, their general application scenarios, and their standard profiles. Furthermore, biometric functions such as enrollment, verification, positive identification, negative identification, and watchlist identification are described in detail in this document. In addition to it, a biometric reference architecture and data exchange paths are shown. At the end, the image of the functions on generic application classes and biometric basic standards is explained. Based on this document, the proposal of a New Work Item was developed, which will be processed in ISO/SC37 in WG 4 (Working Group on Profiles for Biometric Applications). Template Protection and Usage

Title: JTC 1/SC 37 N 43; US Contribution to JTC 1/SC 37 on Template Protection and Usage

Creating and using biometric templates was addressed by some documents, such as for example X9.84, BioAPI specification, and CBEFF specification. X9.84 focuses specially on the role of data encoding and digital signatures for guaranteeing data protection and template integrity. BioAPI deals with adding application-related data. CBEFF describes general data structures, which are used by BioAPI and X9.84. The document "Template Protection and Usage" includes questions of confidentiality, integrity of templates and their respective user data. The aspects of storing data with regard to using templates are becoming more and more important, since more and more applications are developed which are using biometric templates. Guaranteeing confidentiality and template integrity shall prevent "identity theft" and "repetition of template"-attacks. In this document, three scenarios for storing and guaranteeing template safety are shown. Then, problems are taken into consideration, which arise if user data (such as e.g. demographic data) are stored together with the template. Encoding and signatures as protective measures are described.



American National Standard for Information Systems – Data Format for the

Interchange of Fingerprint, Facial, Scar Mark & Tattoo (SMT) Information

ANSI/NIST-ITL 1-2000 Revision of ANSI/NIST-CSL 1-1993 & ANSI/NIST-ITL 1a-1997 For federal and regional criminal prosecution authorities methods which support the identification of persons, based on existing fingerprints and photographs of scars and tattoos, are interesting. In order to guarantee the cooperation between different authorities in the search for criminals, a general standard format for data exchange purposes is necessary. The Information Technology Laboratory (ITL) of the National Institute of Standards and Technology (NIST) supported the development of the document entitled “Data Format for the Interchange of Fingerprint, Facial, Scar Mark & Tattoo (SMT) Information”. This document is an updated standard which substitutes the ANSI/NIST-CSL 1-993 and ANSI/NIST-ITL 1a-1997 standards. It addresses the topic of exchanging fingerprint, facial and scar marks as well as tattoo data. The standard defines content, format, and units of measurement for exchanging data about fingerprints, photographs, and images of scars and tattoos, which can be used for human identification purposes. It describes a number of necessary and optional data elements. The standard does not deal with the characteristics of software systems, which are required for formatting or compressing pictures. This document is quite large. It consists of 24 sections and six appendices. The first three sections are dedicated to "administrative" topics – (firstly) goals and application areas for the standard, and (secondly) further standards which are relevant to the standard, and (thirdly) definitions of the terms and abbreviations used. The fourth section defines conventions for transferring data, such as e.g. • In which order shall bytes be transferred? • How shall compressed and not-compressed grayscale images be transferred? The fifth section describes the requirements for image resolution of fingerprints and of the palm of the hand. This deals with scanner resolutions and transfer resolution while the latter does not necessarily have to be identical with the scanner resolution. The sixth section is a file description. In this section, the composition of a file, which is to be transferred to an authority, is shown. Some parts of this file should meet the requirements of the receiving authority. It should contain one or more logical data records, while each set of data should be one of the fixed types (these different types of data records are defined in Sections 8 to 24). All logical data records, which belong to a transaction, should be part of a physical file. The standard defines (1) three logical data records in order to interchange ASCII text fields, (2) six logical data records to exchange binary data, and (3) another five types of data records in order to exchange a combination of ASCII and image data within a logical memory structure. The seventh section contains a description of the data records. Firstly, types of logical data records are briefly shown. Then, the format for the data records is described. As mentioned before, the sections that follow thereafter are detailed descriptions of the data record types. Six appendices belong to this document. Appendix A is normative and contains 7-Bit ANSI codes for exchanging data. Appendix B is informative and shows the use of dashes for separating data. Appendix C is normative and describes the encoding pattern based on 64-bit. Appendix D is normative and describes JFIF (JPEG File Interchange Format). Appendix E contains excerpts of the 8th version of the "NCIC Code Manual" for the description of scars, marks, and tattoos. Appendix F is an extensive example for a file that contains data records of a fingerprint image, a criminal face photograph, and the palm of a hand. The data records in Appendix F are formatted in accordance with the standard.



Exchange Formats for Biometric Data. Part 1:

General Aspects and Requirements

As Part 1 of the 19794 Standard, a document is planned which, as a framework, contains such aspects, which would be repeated in the next parts 19794-2 through to 19794-6. For example, statements about the storage format and compressing images for Standard No. 19794-4 (finger image) as well as for No. 19794-5 (face image) and for 19794-6 (iris image) are interesting. The plan is for these aspects to be extracted from the individual parts in order to be summarized in the framework document No. 19794-4. The document consists of ten sections. The first four sections describe the application area of the document, the standards that were used, and the relevant terms and abbreviations. General problems, that occur when using biometric data, are laid out in the fifth section. Among others, the natural variableness of biometric data is one of these problems. Variability means that in an authentication process, reference data and authentication data are not identical. The properties of biometric minutiae lead to the fact that one and the same person is not in the position to provide identical data several times. Hence, the same data is never used for authentication purposes but rather more or less similar data. Very often, the mathematical basics for these facts are lacking. Another problem is that some biometric features, such as one's signature and face, are subject to ageing. It is recommended that, for face recognition purposes, no data is used that is "older" than six years. Furthermore, ethnological differences are a challenge for biometric systems which have to be taken into consideration for international projects. The last problem to be mentioned are the algorithms for the extraction of features from the biometric data and for matching biometric data. Different characteristics of the same biometric feature can be extracted and comparisons carried out in different ways. In order to be able to standardize data formats, it would be necessary to examine these algorithms and to identify those that are best suited. The sixth section is entitled "Types of Formats for Exchanging Data". Different types of data are described by briefly explaining how the raw data of the sensor is transformed step by step into data for identification/verification, with the result that different types of data will be created. The seventh section is dedicated to the topic "Nestling of Biometric Data in Other Data Structures”. By defining a number of standard elements which may construct the biometric data, the inter-operability and the exchange of data can be achieved, even if - due to differences in the encoding pattern of the elements - a translation is required. Thanks to the framework for general exchange formats for biometric data (Common Biometric Exchange Formats Framework, CBEFF), this is possible. The same data can exist in different CBEFF-compatible formats which are required in practical applications. However, each field of application can use the data from other fields after format transformation. In addition, CBEFF data structures, BioAPI data structures and ISO/IEC 7816-11 data structures are described briefly. In the eighth section, the conventions for biometric data are dealt with. This section contains such aspects as order of bits and bytes, elements of the header, quality of biometric data (the classes are: "not suitable", "sufficient", "suitable", "very good") as well as safety of biometric data. The ninth section briefly describes some conditions and requirements for the acquisition process of biometric data. It should be a safe and trustworthy authority where the biometric features of a user ought to be acquired. There could be people whose characteristics cannot be recorded. Next to the features, further data such as time of acquisition, memory capacity, etc. can also be incorporated. The final section deals with the topics of feature extraction as well as with match algorithms. First, such areas are specified where standardization activities can be carried out. Among them are: • Data formats

The technical inter-operability of different hardware and software components is made possible through standardization of data formats.



• Algorithms for extracting biometric characteristics Clear correlation between raw data (coming from the sensor) and pattern data (data, which was derived from raw data and only relevant data, such as minutiae for fingerprint images) can be established through standardization.

• Matching algorithms This standardization provides decision criteria for the matching process of biometric.

Such aspects as the formats for pattern data, recognition performance of the algorithms, problems of algorithms for feature extraction purposes, statistic safety levels, etc. are addressed. It is recommended to install a test meter which counts the number of authentication trials and limits them in order to avoid misuse of the biometric systems. Finally, the parameters for matching algorithms are discussed. For smart card matching algorithms these include e.g. • Minimum and maximum number of data elements that are expected by a matching algorithm, • Naming the system that was used for representing the data: Cartesian system, polar system, etc.,

and • Naming a file system for the data such as "x-coordinates in descending order”. Up until now, the document at hand was only used in DIN. There are plans to compile a Committee Draft (CD) in June 2004.

7.2.2 Fingerprint-Specific Standards

DIN V 66400

Finger Minutiae Encoding Format and Parameters for On-Card Matching Title: DIN V 66400 Finger Minutiae Encoding Format and Parameters for On-Card Matching, Version 0.8 DIN V 66400 specifies a format for encoding fingerprint properties (in the form of minutiae). The aim in developing this standard was the encoding of minutiae for verification purposes – especially for comparisons with data stored on a card. The development of applications with high security requirements is made possible through defined encoding and matching parameters. The standardized encoding provides inter-operability between different card and service systems. The standard addresses the verification process only. First, the standards used for the specification are listed and relevant terms defined. Then, minutiae are described: Which characteristics do they have? Which types of minutiae exist? How are they to be localized on the fingerprint image? The next section shows the data format of the minutiae while one has to always take into consideration that the Standard DIN V 66400 addresses verification only. This document specifies standardized biometric data for showing fingerprints in the form of minutiae for which the combination of standardized data with proprietary data is permissible. The structure of the minutiae-based standard data is described for fingerprint image verification. In Standard DIN V 66400, two formats for a minutiae-based presentation of fingerprint images are defined: normal format and compact format. As for the normal format, 5 bytes are allocated per minutia. In the compact format minutiae only 3 bytes are used which leads to a reduction in storage costs. Further aspects, which are dealt with in the document, are number, order, and reduction of the minutiae. The number of minutiae which conforms to Standard DIN V 66400 depends on the desired security level and the specific card. The number of minutiae as well as their order is specific to implementation – these aspects are determined by the use of actual cards. If the number of minutiae



that was delivered exceeds the maximum number of minutiae that can be accepted by a card, then a minutiae reduction has to be carried out. Another section describes parameters that are relevant to (1) enrollment (acceptance of fingerprints), (2) for comparison and (3) verification purposes, such as e.g. the number of minutiae, matching conditions, and the number of matching trials. Afterwards, the presentation of biometric data is shown. It contains information about data required for verification and a description of the matching algorithm of the card. Among other things, it contains the following data: • Type of biometric technology (in this case "fingerprint") • Instance of the biometric type (e.g. "left index finger") • Holder of the format (Confer CBEFF) • Format type, and • Parameter of the matching algorithm DIN V 66400 was introduced as the German contribution to the SC37 WG3 Special Group on Biometric Data Interchange Formats. The aim is to introduce this standard within the work of the Finger Minutiae Format for Data Interchange. Finger Minutiae Data

Title: Biometric Data Interchange Formats – Part 2: Finger Minutiae Data This standard defines a method to represent fingerprint image data. Its method is based on the minutiae principle. The standard specifies the concept and data format for showing fingerprint images in the form of minutiae. The data format is generic. Hence, it can be used in many applications with automatic fingerprint recognition. This standard deals with such requirements or characteristics that are not application-specific. Devices and algorithms which are proving to conform to the "Finger Minutiae Format for Data Interchange" Standard shall fulfill the defined requirements. The requirements refer, among other things, to order and size of the data fields, existence of all necessary data fields, consideration of value ranges for data, and internal consistency (e.g. the number of data records of individual fingerprint images should correspond with the number of fingers in the respective data field). The standard is based on the “American Association of Motor Vehicle Administrators Driver License Standard 2000” (AAMVA Standard 20000630). As a biometric standard for data formats, it has a CBEFF format holder and a format type code. It contains the definitions of all relevant terms. One section deals with the topic "Extraction of Minutiae" since it is significant to the inter-operability that the same method and the same procedure are always used for extracting minutiae. Another section defines the format for storing minutiae. The Draft for the Standard Finger Minutiae Format for Data Interchange is processed with ISO/SC37 in WG 3 (Working Group on Biometric Data Interchange Formats) as CD 19794-2.



Finger Image Data

Title: Biometric Data Interchange Formats – Part 4: Finger Image Data

This standard draft can be used for interchanging and matching fingerprint images. It focuses on such applications that have no limits with regard to memory space. While other standards are about interchanging lists of different finger characteristics such as e.g. minutiae or patterns, the standard at hand deals with exchanging complete fingerprint images. Only if certain fingerprint characteristics are acquired and transmitted in a standard format, less memory space will be required for the acquisition and transmission of entire fingerprint images. On the other hand, data about minutiae cannot be used in a pattern recognition algorithm. However, if a complete fingerprint image exists, it can be used by all algorithms independent of how they function and independent of their manufacturer. There are different algorithms in the fingerprint image recognition technology. Independent of how they function, a fingerprint of a relatively high quality should be obtained first. The standard at hand specifies formats for interchanging data records for storage purposes and for obtaining and transferring data about fingerprint images. It defines content, format, and units for interchanging fingerprint images, which are used for identification and verification purposes. This data consists of a large number of required and optional parameters, such as e.g. parameters with regard to finger scanning, compressed and non-compressed images as well as data specific to each manufacturer. Two aspects are dealt with in the section entitled "Data conventions": Bytes and bits order and scanning sequences. The ordering of bytes and bits determines how fingerprint images are to be stored. The scanning aspect deals with scanning the fingers, e.g. with the orientation of the fingers, their placement at the beginning of the axis (coordinates 0, 0), etc. Another section is entitled "Requirements for fingerprint images”. Four fingerprint image quality classes are defined. Parameters such as pixel depth, certification, and scanning resolution, etc. are selected. Afterwards the values of these parameters, which have to be maintained are determined for every quality class. The last section is entitled "Format of fingerprint data records”. This section defines of which elements a fingerprint data record consists: What are the elements called? Which format do they have? How many bytes shall be used for each element? The draft for this standard is processed in ISO/SC37 in WG 3 (Working Group on Biometric Data Interchange Formats) as CD 19794-4.



Finger Pattern Data

Title: Biometric Data Interchange Formats – Part 3: Finger Pattern Data

This standard draft specifies the exchange format for interchanging fingerprint images, which are suitable for pattern recognition algorithms. Pattern-based algorithms process "global" parts of the fingerprint images while feature-based algorithms extract specific characteristics from the images. Currently, there are no established mechanisms for exchanging fingerprint pattern data that are used by a pattern-based algorithm. This standard starts with a description of all relevant terms and the used standards. Then, the topic "finger pattern data" is addressed. It describes how the obtained fingerprint images are processed. Since pattern-based algorithms need a relatively low resolution, finger images with a low resolution are obtained from the original images in a first step. Secondly, a presentation of the fingerprint images is generated with which the actual finger pattern data for the exchange is established. Then, data records for pattern-based fingerprint data are described. The data record format contains both general as well as individual data elements (specific to each manufacturer). Apart from the format identifier and the number of the version, all other data elements are to be binary. There are no separators. As a standardized biometric data format, this format has a CBEFF format holder and format type codes. Thereafter, the exact composition of a data record is described. In the Appendix, an example of a data record for pattern-based fingerprint image data is shown. The draft for this standard is processed in ISO/SC37 in WG 3 (Working Group on Biometric Data Interchange Formats) as CD 19794-3.



8 Bibliography [BEM02] Biometric Evaluation Methodology Working Group – Common Methodology for

Information Technology Security Evaluation, Version 1.0, August 2002 [BIOIS] Studie BioIS, Feldversuch am Fraunhofer-IGD,

http://www.igd.fhg.de/igd-a8/projects/biois [BP02] Mansfield, A.J., NPL, Wayman, J.L., SJSU, NPL: Best Practices in Testing

and Reporting Performance of Biometric Devices, Report CMSC 14/02, Version 2.01, August 2002

[BPG02] Appendix G: Interim IAFIS Image Quality Specifications for Scanners.

Electronic Fingerprint Transmission Specification, Criminal Justice Information Services, CJIS-RS-0010 (V4), 1995

[BRO02] Bromba, M.: BioIdentifikation, Siemens, 2002 [BSI00] BSI, Technische Evaluierungskriterien zur Bewertung und Klassifizierung

biometrischer Systeme, October 2000 [CT02] c‘t - Magazin für Computertechnik, 11/02, pages 114-123 [DRA01] Drahansky, M.: Fingerabdruckerkennung mittels neuronaler Netze, Diploma

thesis, Open University in Hagen; VUT in Brno, 2001 [FVC00] Maio, D., Maltoni, D., Cappelli, R., Wayman, J.L., Jain, A.K.: FVC2000 – Fingerprint

Verification Competition, University of Bologna & San Jose State University & Michigan State University, 2000

[FVC02] Maio, D., Maltoni, D., Cappelli, R., Wayman, J.L., Jain, A.K.: FVC2000 – Fingerprint

Verification Competition, University of Bologna & San Jose State University & Michigan State University, 2000

[GRI] Griesser, H.: Biometrie, Biometrix Int., Vienna-Austria [HEI27] Heindl, R.: System und Praxis der Daktyloskopie und der sonstigen

technischen Methoden der Kriminalpolizei, Berlin and Leipzig, 1927 [HEN03] The Henry Classification System, International Biometric Group, New York,

USA, 2003 [IBG03] International Biometric Group: Draft Project Proposal – Evaluating Multi-Modal

Biometric Systems, 2003 [MAL03] Maltoni, D., Maio, D., Jain, A.K., Prabhakar, S.: Handbook of Fingerprint

Recognition, 2003 [NORM0] DIN NI-AHGB N 0086; NI-17 N 0640; NI-37 N 0003


http://www.igd.fhg.de/igd-a8/projects/biois


ISO/IEC JTC 1/SC 37 N55 2002-12-17 ISO/IEC 19784 Date: 2002-12-13 Ref. number: ISO/IEC JTC 1/SC 37 N 55 [NORM1] DIN NI-37-N0108, fer. DIN NI-AHGB N 0087; NI-17 N 0641; NI-37 N 0005 ISO/IEC JTC 1/SC 37 N56 2002-12-17 [NORM2] ISO/IEC JTC 1/SC 37 N92 2003-03-18 Moreover, ISO/IEC JTC 1/SC37 N125 and N197 [NORM3] DIN NI-AHGB N 0069 ISO/IEC JTC 1/SC 37 N43 2002-11-06 ISO/IEC JTC 1/SC 37 [NORM4] ANSI/NIST-ITL 1-2000 Revision of ANSI/NIST-CSL 1-1993 & ANSI/NIST-ITL 1a-1997 [NORM5] DIN, NI-AHGB N 0014 [NORM6] ISO/IEC JTC 1/SC 37 N184, as well as ISO/IEC JTC 1/SC 37 N90 2003-03-18 [NORM7] ISO/IEC JTC 1/SC 37 N0180 ISO/IEC JTC 1/SC 37 N89 2003-03-18 [NORM8] ISO/IEC JTC 1/SC 37 N183 2003-04-16 WD 19794-3, as well as ISO/IEC JTC 1/SC 37 N91 2003-03-18 [NORM9] ISO/IEC JTC 1/SC 37, Date: 2003-06-23 Proposal on ISO/IEC WD 19794-1 ISO/IEC JTC 1/SC 37/SG 3, Secretariat: DIN [RAT02] Ratha, N.K., Senior, A., Bolle, R.M.: Automated Biometrics, 2002 [STR02] Struif, B., Müller, R.: DIN V 66400, 2002 [TTT02] TeleTrusT Deutschland e.V.: Bewertungskriterien zur Vergleichbarkeit biometrischer

Verfahren, 2002 [WET02] Wetting, S.: Biometrie: Verfahren und ausgewählte Rechtsprobleme,

Friedrich-Schiller-University Jena, 2002 [ZAM91] Zamperoni, P.: Methoden der digitalen Bildsignalverarbeitung, Vieweg, 1991 [ZWI00] Zwiesele, A., Munde, A., Busch, C., Daum, H.: "Comparative Study of

Biometric Identification Systems" in: 34th Annual 2000 IEEE International Carnahan Conference on Security Technology, page 60

Websites of respective German and international companies.



9 Table of Abbreviations

AFIS Automated Fingerprint Identification System

ATV Ability To Verify

CCD Charge-Coupled Device

DB Database

DET Detection Error Trade-Off

EER Equal-Error Rate

FA Fingerprint

FAES Fingerprint Acceptance System

FAR False Acceptance Rate

FMR False Match Rate

FNMR False Non-Match Rate

FP Fingerprint (in German)

FRR False Rejection Rate

FTA Failure To Acquire

FTC Failure To Capture

FTE Failure To Enroll

FTM Failure To Match

GD Genuine Distribution

ID Impostor Distribution

LED Light Emitting Diode

ME Minutiae or Feature Acceptance or Extraction

MEi Feature Extraction Procedure Number j

MoC Match On Card

MSA Matching Score Algorithm

MSAk Matching Score Algorithm Number k

MTBF Mean Time Between Failures, ISO/DIN 40042

NME Number of Feature Extraction Procedures

NMSA Number of Matching Score Algorithms

NS Number of Sensors

ROC Receiver Operating Characteristic

Si Sensor Number i

SNR Signal-to-Noise-Ratio


evaluation of fingerprint recognition

Documents