(spring 2013) signature authentication consistency across devices

1
SIGNATURE AUTHENTICATION CONSISTENCY ACROSS DEVICES The purpose of this study is to determine the consistency of signatures and the reliability of verification across various devices. These devices include an iPad (with finger and stylus) and digitizers (back-lit and non-back-lit) for the purpose of electronic consent. This study was conducted using two to four (varied for each device and person) signatures per team member and then cross-comparing every combination of signature to determine the stability of using the algorithm across various devices with a given threshold. David Mihai, Jasmine Jones, Michael Brockly, Richard Guest, Stephen Elliott Overview Devices Used Matlab SOP Results 73 87 18 12 0 20 40 60 80 100 120 Number of Comparisons Results 38% 10% 46% 6% Signature Results (Uncorrected) Positive False Positive Negative False Negative 74 91 17 8 0 20 40 60 80 100 120 Number of Comparisons Results 39% 9% 48% 4% Signature Results (Corrected for Image Type) Positive False Positive Negative False Negative From the results, and based off of the assumption that any margin of error greater than 3% was considered to be unsatisfactory, the devices used could not reliably compare signatures using the algorithm. “Positive” means we matched a valid signature across devices. “False Positive” means that a signature was recognized as valid when it was not. “Negative” means that a signature was recognized to be invalid at the proper time. “False Negative” is when a signature that was valid was recognized as an invalid signature. This analysis includes results that were both corrected and uncorrected for image type and quality (such as resolution or sharpness of image). This correction accounts for the differences between, in this particular experiment, lower-quality bitmap files and higher-quality jpg images. In order to conduct this experiment, a tool was needed to compare and provide a quantitative value of signature similarity. In order to process this data, an algorithm provided by Dr. Guest at the University of Kent was used. The SURF standard signature algorithm returns a series of mean reference point distances between greyscale images. In order to correct for outliers and to have the most accurate results, only the nearest 50% of distances were used in calculating the sum. For example, using this algorithm, calling: >> SURFstaticcompare(‘image1.bmp’,’image2.bmp') Using this algorithm in Matlab 7.8.0 will return – ans = sum1: 0.1363 (Returns 75% of points) sum2: 0.1295 (Returns 66% of points) sum3: 0.1198 (Returns 50% of points – value used as per Dr. Guest’s recommendation) sum4: 0.1084 (Returns 33% of points) sum5: 0.1647 (Returns 100% of points) The sum3 values were then analyzed and compared to a threshold to determine a match or rejection dependent on the image type and quality. iPad 2 ePad ink backlit Signature Application on iPad Stylus used with iPad ePad ink Non-Backlit

Upload: international-center-for-biometric-research

Post on 28-Nov-2014

704 views

Category:

Technology


0 download

DESCRIPTION

The purpose of this study is to determine the consistency of signatures and the reliability of verification across various devices. These devices include an iPad (with finger and stylus) and digitizers (back-lit and non-back-lit) for the purpose of electronic consent. This study was conducted using two to four (varied for each device and person) signatures per team member and then cross-comparing every combination of signature to determine the stability of using the algorithm across various devices with a given threshold.

TRANSCRIPT

Page 1: (Spring 2013) Signature Authentication Consistency Across Devices

SIGNATURE AUTHENTICATION CONSISTENCY ACROSS DEVICES

The purpose of this study is to determine the consistency of signatures and the reliability of verification across various devices. These devices include an iPad (with finger and stylus) and digitizers (back-lit and non-back-lit) for the purpose of electronic consent. This study was conducted using two to four (varied for each device and person) signatures per team member and then cross-comparing every combination of signature to determine the stability of using the algorithm across various devices with a given threshold.

David Mihai, Jasmine Jones, Michael Brockly, Richard Guest, Stephen Elliott

Overview

Devices Used Matlab SOP

Results

73

87

18 12

0

20

40

60

80

100

120

Num

ber o

f Com

paris

ons

Results

38%

10%

46%

6%

Signature Results (Uncorrected)

Positive

FalsePositive

Negative

FalseNegative

74 91

17 8

0

20

40

60

80

100

120

Num

ber o

f Com

paris

ons

Results

39%

9%

48%

4%

Signature Results (Corrected for Image

Type)

Positive

FalsePositive

Negative

FalseNegative

From the results, and based off of the assumption that any margin of error greater than 3% was considered to be unsatisfactory, the devices used could not reliably compare signatures using the algorithm. “Positive” means we matched a valid signature across devices. “False Positive” means that a signature was recognized as valid when it was not. “Negative” means that a signature was recognized to be invalid at the proper time. “False Negative” is when a signature that was valid was recognized as an invalid signature. This analysis includes results that were both corrected and uncorrected for image type and quality (such as resolution or sharpness of image). This correction accounts for the differences between, in this particular experiment, lower-quality bitmap files and higher-quality jpg images.

In order to conduct this experiment, a tool was needed to compare and provide a quantitative value of signature similarity. In order to process this data, an algorithm provided by Dr. Guest at the University of Kent was used. The SURF standard signature algorithm returns a series of mean reference point distances between greyscale images. In order to correct for outliers and to have the most accurate results, only the nearest 50% of distances were used in calculating the sum. For example, using this algorithm, calling: >> SURFstaticcompare(‘image1.bmp’,’image2.bmp') Using this algorithm in Matlab 7.8.0 will return – ans = sum1: 0.1363 (Returns 75% of points) sum2: 0.1295 (Returns 66% of points) sum3: 0.1198 (Returns 50% of points – value used as per Dr. Guest’s recommendation) sum4: 0.1084 (Returns 33% of points) sum5: 0.1647 (Returns 100% of points) The sum3 values were then analyzed and compared to a threshold to determine a match or rejection dependent on the image type and quality.

iPad 2 ePad ink backlit

Signature Application on iPad Stylus used with iPad

ePad ink Non-Backlit