[communications in computer and information science] advances in digital image processing and...

10
D. Nagamalai, E. Renault, and M. Dhanushkodi (Eds.): DPPR 2011, CCIS 205, pp. 146–155, 2011. © Springer-Verlag Berlin Heidelberg 2011 Automatic, Robust Face Detection and Recognition System for Surveillanceand Security Using LabVIEW (sCUBE) C. Lakshmi Deepika, M. Alagappan, A. Kandaswamy, H. Wassim Ferose, and R. Arun Department of Biomedical engineering, PSG College of Technology, Coimbatore -641004, India Abstract. The automatic, high end surveillance systems are of immense need in the wake of the emerging security problems faced in today’s world. Most of the high end systems use current trends in technology but often prove to be costly which make them un-affordable for the common people. Thus there is an urge to develop a fully functional, high end, continuous surveillance system which has an error free monitoring and also cost effective. Thus we have taken up the challenge of developing a low cost, real time face detection and face recognition system which can provide automatic, robust, unmanned Surveillance and Security at critical points. The system was successfully installed and the efficiency of the overall system was tested. 1 Introduction The application we have developed is a Security System, which uses Face Extraction and Recognition as its underlying principle. Face is one of the qualifying biometric identifier, which can be used to identify an individual effectively. Since the face biometric possesses the merits of high accuracy and low intrusiveness, it has drawn attention from various researchers especially in areas of security. An automatic system is very useful when due to the inherent nature of a human operator such as boredom or reduced alertness levels; he is not able to effectively scrutinize the voluminous videotape produced by the surveillance cameras. This proposed system can be used to make the security at the important places more effective in cases such as authenticating authorized persons, identifying in advance the visiting of VIPs, for detection of a criminal, for protection of undesirable property loss and to save valuable lives by defending or guarding from attack. We hence propose to solve the challenge by providing visual facility to a computer and convert it to a multi functionality system which will continuously monitor a given critical entry point or area using a high resolution camera. The system will be able to detect human face from the camera output, recognize it from a database and allow the person if authorized or indicate an emergency if unauthorized. The image processing algorithms we have developed for face detection and recognition are computationally

Upload: murugan

Post on 23-Dec-2016

213 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: [Communications in Computer and Information Science] Advances in Digital Image Processing and Information Technology Volume 205 || Automatic, Robust Face Detection and Recognition

D. Nagamalai, E. Renault, and M. Dhanushkodi (Eds.): DPPR 2011, CCIS 205, pp. 146–155, 2011. © Springer-Verlag Berlin Heidelberg 2011

Automatic, Robust Face Detection and Recognition System for Surveillance and Security Using

LabVIEW (sCUBE)

C. Lakshmi Deepika, M. Alagappan, A. Kandaswamy, H. Wassim Ferose, and R. Arun

Department of Biomedical engineering, PSG College of Technology, Coimbatore -641004, India

Abstract. The automatic, high end surveillance systems are of immense need in the wake of the emerging security problems faced in today’s world. Most of the high end systems use current trends in technology but often prove to be costly which make them un-affordable for the common people. Thus there is an urge to develop a fully functional, high end, continuous surveillance system which has an error free monitoring and also cost effective. Thus we have taken up the challenge of developing a low cost, real time face detection and face recognition system which can provide automatic, robust, unmanned Surveillance and Security at critical points. The system was successfully installed and the efficiency of the overall system was tested.

1 Introduction

The application we have developed is a Security System, which uses Face Extraction and Recognition as its underlying principle. Face is one of the qualifying biometric identifier, which can be used to identify an individual effectively. Since the face biometric possesses the merits of high accuracy and low intrusiveness, it has drawn attention from various researchers especially in areas of security. An automatic system is very useful when due to the inherent nature of a human operator such as boredom or reduced alertness levels; he is not able to effectively scrutinize the voluminous videotape produced by the surveillance cameras. This proposed system can be used to make the security at the important places more effective in cases such as authenticating authorized persons, identifying in advance the visiting of VIPs, for detection of a criminal, for protection of undesirable property loss and to save valuable lives by defending or guarding from attack.

We hence propose to solve the challenge by providing visual facility to a computer and convert it to a multi functionality system which will continuously monitor a given critical entry point or area using a high resolution camera. The system will be able to detect human face from the camera output, recognize it from a database and allow the person if authorized or indicate an emergency if unauthorized. The image processing algorithms we have developed for face detection and recognition are computationally

Page 2: [Communications in Computer and Information Science] Advances in Digital Image Processing and Information Technology Volume 205 || Automatic, Robust Face Detection and Recognition

Automatic, Robust Face Detection and Recognition System 147

complex and require a dynamic image processing tool to adapt it. The Vision Assistant, DAQ Assistant and Serial Port Communication in LabVIEW made it possible to successfully implement our idea in the proposed system.

The graphical system design of our system enhances the programming ability of the programmers. Its versatile design and the application enable the maximum dynamism in developing the complex application such as ours, in a better way rather than the text based programming languages. It reduces the development time of applications as well as increases the interest of developing new algorithms. The utmost user-friendly structures creates an environment to identify the application easily even by the idle man. The attractive visualization of the tools induces new innovations in the design during development and enriches the effective handling capability in the real time implementations.

There are several approaches to face detection and recognition in literature. Kanade presented an automatic feature extraction method based on the ratios of distances between facial features and reported a recognition rate of about 45-75% with a database of 20 people [1]. Brunelli and Poggio compute a set of geometrical features such as nose width and length, mouth position and chin shape. They were able to obtain 90% recognition with a database of 40 people [2]. Turk and Pentland proposed a face recognition scheme in which the face images are projected onto the principal components of the original set of training images [3]. This method also called Principal Component Analysis (PCA) later became a defacto standard in the field of pattern recognition and also as benchmark for testing new methods in research on pattern recognition. We propose to use PCA in our real-time system.

2 Image Acquisition

A camera is mounted at a vantage point at the entrance of the area under surveillance. The entry of a person into the area is detected by using a laser beam. When the laser

Fig. 1. Steps involved in Image Acquisition

Page 3: [Communications in Computer and Information Science] Advances in Digital Image Processing and Information Technology Volume 205 || Automatic, Robust Face Detection and Recognition

148 C.L. Deepika et al.

beam is cut due to the person crossing by, it produces a software trigger to the camera, which creates a snapshot of the area under its focus. This photograph will have the person as well as the background and any other objects in it. It is transferred to the PC in which our LabVIEW application is installed. The Real time Image Acquisition facility in LabVIEW package helped us to simultaneously process the photograph as and when the person was captured in the area under surveillance.

3 Face Detection

This can also be called Face Segmentation where the facial images have to be segmented from the background. This task is seemingly effortless when it is done

Fig. 2. Steps involved in Face Detection

Page 4: [Communications in Computer and Information Science] Advances in Digital Image Processing and Information Technology Volume 205 || Automatic, Robust Face Detection and Recognition

Automatic, Robust Face Detection and Recognition System 149

by a human; however the underlying computations within t he human visual systems are very complex. A full-fledged Surveillance and Security system has to detect locate and segregate faces, which will then be given as input to the Face Recognition system. According to the literature survey, it is found that the different human skin colors from different races, falls in a compact region of the color spaces. Hence, the image obtained from the camera is first transformed to the color space model namely the YCbCr model which gives the best results for skin pixel detection [4]. The characteristics of the face can be changed due to unexpected conditions such as shadows or change in lighting. Hence the lighting compensation is provided, the skin regions are extracted and the noise components are removed. Next the skin regions are labeled and identified and the labeled regions are tested for face criteria. Finally the possible faces are extracted.

Hence the automation of this task is a complex one, but the Vision Development Module of LabVIEW package with its vast library of powerful operators helped us to implement our complex algorithms magnificently.

4 Face Recognition

It is basically a machine learning or pattern classification problem. The extracted face image is converted to a set of features called the feature vector and compared with the facial database. This is also a quite challenging task since the identification of the face image has to be done in spite of variations due to variable illumination, facial expression and decoration (eg., glasses). It may also be required to identify the test face image from a large gallery of facial images, which is also a skilful development task.

To perform the face recognition, here a statistical tool namely Principal Component Analysis is used [5-7]. The power of PCA lies in its ability to convert a large dimensional data space (the image database) to a smaller dimensional feature space, so that the presence or absence of the test image in the database can be found with less number of computations. In this method, every face image, which is a 2D matrix, is converted into 1D arrays each of dimension N. The M arrays corresponding to M face images, that are arranged as a 2D array to form the data set and the empirical mean of all the N columns are calculated. The mean centered 2D array is formed by finding the deviation of every column from its mean. The Covariance matrix is then constructed. The Eigenvectors and Eigen values of the matrix are found and arranged according to their values. The largest eigen values are the principal components or eigen faces, which forms the basis for the large image space. The input test face is projected into this eigen face space and using Euclidean Distance metric, the nearest matching face image is retrieved from the database

Page 5: [Communications in Computer and Information Science] Advances in Digital Image Processing and Information Technology Volume 205 || Automatic, Robust Face Detection and Recognition

150 C.L. Deepika et al.

Fig. 3. Steps involved in Face Detection

5 Implementation Using LabVIEW

The implementation of the face recognition and detection system is explained as four modules:

A. Image Acquisition The camera used here is an NI Compatible Scout camera and is triggered by the sensor that is mounted in front of the door, which is activated during the nearby visit of the person. Then the immediate image taken by the camera is processed by the face detection section.

B. Face Detection • The captured RGB image is converted into YCbCr color space model

using user defined functions.

Page 6: [Communications in Computer and Information Science] Advances in Digital Image Processing and Information Technology Volume 205 || Automatic, Robust Face Detection and Recognition

Automatic, Robust Face Detection and Recognition System 151

• The lighting compensation is carried on to the test image using statistical formulas to isolate the background color.

• The skin regions are extracted using functions available in Vision and Motion tools.

• The noise is removed from the extracted skin region using the smoothing filter.

• The skin color box is identified using Blob analysis. • The face criteria are implemented using the functions available in Vision

and Motion tools and programming tools. • The possible face region is detected after checking the face criteria.

C. Face Recognition • Every face image which is in the database is converted into a 1D array

of N elements where N is the number of pixels in the image. Every array corresponding to single images are then appended to form a 2D array to form the dataset.

• The empirical mean of every column is found. • The deviation of every column from its mean is then found.

The covariance matrix is then constructed. • The eigen values and eigenvectors of this matrix are calculated and

sorted. • The largest eigen values are taken to form the basis of the eigen face space. • The test face is now projected to this eigen face space. • The nearest matching image is retrieved from the database using

Euclidean distance metric which gives the image of the person passing by.

All the above steps are performed using the functions available in Programming tools, Mathematics tools, and Vision and Motion tools.

D. Database Management Here are some of the features which we have implemented in sCUBE using functions available in programming tools.

• View Database – To view the complete list of authorized visitors and their personal details.

• Appending to Database – To add the new record into the database. • Remove from Database - To remove the specific entry from the

database. • Specific Search – To search for a specific visitor and his visiting

times. • General Search – To search for all the visitors and their visiting times. • View Sent Mail - To view the list of E-Mails sent during the visit o f

unauthorised persons. • Maintenance of image database for unauthorized visitors.

Page 7: [Communications in Computer and Information Science] Advances in Digital Image Processing and Information Technology Volume 205 || Automatic, Robust Face Detection and Recognition

152 C.L. Deepika et al.

6 Results and Discussion

A. Face Detection

Fig. 4. Face Detection Performed on a test image taken using our camera

Fig. 5. Face Detection Performed on a test image available in Internet database

B. Face Recognition

Fig. 6. Face Recognition performed for the corresponding Detected Image

Fig. 7. Face Recognition performed for the corresponding Detected Image

Page 8: [Communications in Computer and Information Science] Advances in Digital Image Processing and Information Technology Volume 205 || Automatic, Robust Face Detection and Recognition

Automatic, Robust Face Detection and Recognition System 153

To verify the performance of our face recognition module we have used freely downloadable databases from the internet (NLPR database with 450 images, AR database with 4000 images, ORL database with 400 images), we have also created our own database with 200 images similar to the above public database, under conditions of varying pose, facial expression and illumination. The results we obtained by using the ORL database is given in Table-1:

S.No Number of Principal

Components Recognition Rate (using Euclidean Distance)

1 50 73.27 2 60 73.59 3 70 73.95 4 80 73.84

We were able to obtain an accuracy rate of 73.95%. However we were able to

obtain about 90.18% accuracy rate when we applied our algorithm to our own database consisting of images with very less variation in pose, expression and illumination compared to the ORL database.

C. Database Management

Fig. 8. Database Management – View Database, Appending to Database, Remove from Database

D. Other Features

• Voice Alert – To welcome the authorized entries. • Alarm – Raised during the entry of unauthorized person . Opening/

Closing of doors – Automated for the authorized entry. • SMS Alert – Sent to the control room during the unauthorized entry.

The following is the front panel of our proposed system

Page 9: [Communications in Computer and Information Science] Advances in Digital Image Processing and Information Technology Volume 205 || Automatic, Robust Face Detection and Recognition

154 C.L. Deepika et al.

Fig. 9. Database Management – General Search, Specific Search, View Sent Mail

Fig. 10. Front panel (1) of the proposed system

Fig. 11. Front panel (2) of the proposed system

Page 10: [Communications in Computer and Information Science] Advances in Digital Image Processing and Information Technology Volume 205 || Automatic, Robust Face Detection and Recognition

Automatic, Robust Face Detection and Recognition System 155

7 Conclusion

We have thus successfully developed the system and tested the its working in the Biomedical Engineering Department at PSG College of Technology. We were able to achieve considerably high performance rate in surveillance using our proposed face detection and recognition methods. The successful implementation of the above automatic, robust face detection and recognition system using LabVIEW proves that with cutting edge technology and powerful software suites available nowadays, it is possible to implement any complex task, provided that the idea and the proposed methodology behind the task are defined very clearly.

References

1. Kanade, T.: Picture Processing by computer complex and recognition of human faces, PhD Thesis, Kyoto University (1973)

2. Brunelli, R., Poggio: Face Recognition: Features versus Templates. IEEE Transactions on Pattern Analysis and Machine Intelligence 15(10), 1042–1052 (1993)

3. Turk, M.A., Pentland, A.P.: Face Recognition Using Eigenfaces. In: IEEE Conf. on Computer Vision and Pattern Recognition, pp. 586–591 (1991)

4. Demers, D., Cotrell, G.W.: Non-Dimensionality reduction. In: Advances in Neural Information Processing Systems, vol. 5, pp. 580–587. Morgan Kaufman Publishers, San Mateo (1993)

5. Turk, M., Pentland, A.: Eigenfaces for Recognition. Journal of Cognitive Neuroscience (March 1991)

6. Turk, M.A., Pentland, A.P.: Face recognition using eigenfaces. In: Proc. of Computer Vision and Pattern Recognition, pp. 586–591. IEEE, Los Alamitos (1991b)

7. Moon, H., Phillips, P.J.: Computational and Performance aspects of PCA-based Face Recognition Algorithms. Perception 30, 303–321 (2001)