System for registering and authenticating human face using support vector machines and method thereof

• US 7,092,555 B2
• Filed: 05/20/2002
• Issued: 08/15/2006
• Est. Priority Date: 12/05/2001
• Status: Expired due to Fees

• Current Assignee: Electronics and Telecommunications Research Institute
• Original Assignee: Electronics and Telecommunications Research Institute
• Inventors: Byun, Hyeran, Park, Chee Hang, Lee, Kyunghee, Chung, Yongwha
• Primary Examiner(s): Wu, Jingge
• Assistant Examiner(s): Tarcu, Robert
• Application Number:
  US10/147,973
• Publication Number:
  US 20030103652A1
• Time in Patent Office: 1,548 Days
• Field of Search: 340/5.3
• US Class Current: 382/118
• CPC Class Codes: G06V 40/172 Classification, e.g. identification

Abstract

A system for performing face registration and authentication using face information, and a method thereof. A set of readily distinguishable features for each user is selected at a registration step and only the set of features selected at the registration step is used at a face authentication step, whereby memory use according to unnecessary information and amount of data calculation for face authentication can be reduced. Thus, the present system has an advantage in that identity authentication through face authentication can be performed even under restricted environments of a USB token or smart card with limited resources. The present system further has advantages in that authentication performance is improved, as readily distinguishable feature information is used, and the time for face authentication is reduced, as face authentication is performed using the SVM built by using the optimal set of readily distinguishable features at a training step.

Claims

1. A face registration system for face authentication using a SVM comprising:

an image input unit for inputting face images of users to be registered in the registration system;

a face and eye region extraction unit for extracting face and eye regions from each face image inputted to the image input unit;

a control unit for classifying the face images of users into training and tuning face images and forming a set of features representing chromosomes of each face image of training and tuning images; and

a SVM classifier for iteratively measuring fitness value of the set of features of each chromosome of the training image using GA (Genetic Algorithms) to select the set of features most suitable to be a feature vector of a corresponding user requesting face authentication and extracting a feature vector from the face image of other users corresponding to the selected set of features, thereby comparing two vectors to determine a result of face authentications;

wherein the eye regions are extracted from the face region through histogram analysis using edge components and binary information.

2. The face registration system as set forth in claim 1, wherein the SVM classifier establishes a set of features with improved fitness value by a chromosome evolution process of crossover and mutation according to GA, as an input vector, upon measuring fitness value of the set of features representing each chromosome.

3. The face registration system as set forth in claim 1, wherein the control unit previously calculates and stores the whole feature values for the training and tuning face images to be used at a training step for selecting the set of features.

4. The face registration system as set forth in claim 1, wherein the control unit normalizes each face image inputted thereto to a predetermined size using eye position information before the calculation of the set of features.

5. The face registration system as set forth in claim 1, wherein the feature values are averages of pixels within each window, the windows being overlapped in a predetermined size to scan each face image.

6. The face registration system as set forth in claim 1, wherein the feature values are averages of edge components of pixels within each window, the windows being overlapped in a predetermined size to scan each face image.

7. The face registration system as set forth in claim 1, wherein the feature values are entropy values of pixels within each window, the windows being overlapped in a predetermined size to scan each face image.

8. A face authentication system using a SVM comprising:

an image input unit for inputting face images of users to be registered in the authentication system;

a face and eye region extraction unit for extracting face and eye regions from each image inputted to the image input unit;

a personal information input unit for inputting personal identification information of a specific user registered in the authentication system to perform face authentication upon receiving a request for identity authentication by the user;

a control unit for reading support vectors, weights and a selected set of readily distinguishable features of the user previously stored, corresponding to personal identification information for face authentication of the user inputted to the personal information input unit, and extracting feature values from the face image inputted to the image input unit corresponding to the set of readily distinguishable features previously stored; and

a SVM classifier for using the support vectors and weights to build the SVM and using an input vector of the feature values extracted from the face image, thereby performing face authentication using the SVM;

wherein the eye regions are extracted from the face region through histogram analysis using edge components and binary information.

9. The face authentication system as set forth in claim 8, wherein the feature values are averages of pixels within each window, the windows being overlapped in a predetermined size to scan each face image.

10. The face authentication system as set forth in claim 8, wherein the feature values are averages of edge components of pixels within each window, the windows being overlapped in a predetermined size to scan each face image.

11. The face authentication system as set forth in claim 8, wherein the feature values are entropy values of pixels within each window, the windows being overlapped in a predetermined size to scan each face image.

12. A face authentication system using a SVM comprising:

an image input unit for inputting face images of users to be registered in the authentication system;

a face and eye region extraction unit for extracting face and eye regions from each image inputted to the image input unit;

a smart card internally having a SVM classifier and to build the SVM using support vectors and weights previously stored for face authentication of the user and receive feature values extracted from a face image of a user requesting identity authentication as an input vector, thereby performing face authentication;

an identification recognition unit for interfacing transmitted/received data between the smart card and authentication system; and

a control unit for inputting values of a set of readily distinguishable features of the user from the smart card, and extracting feature values from the face image of the user corresponding to the set of readily distinguishable features previously stored, thereby transferring an input vector of the extracted feature values to the smart card, thereby requesting face authentication;

wherein the eye regions are extracted from the face region through histogram analysis using edge components and binary information.

13. A method for face registration by a face registration system using a SVM, comprising the steps of:

(a) inputting face images of users registered in the registration system;

(b) extracting face and eye regions from each inputted image;

(c) calculating feature values corresponding to whole sets of features to be used to select features from the inputted image; and

(d) forming a set of features representing each chromosome of the face image and using the set of features having highest fitness value selected through feature set selection step using GA as a feature vector for face authentication of each user, thereby forming and registering a SVM;

wherein the eye regions are extracted from the face region through histogram analysis using edge components and binary information.

14. The method as set forth in claim 13, wherein the step (b) includes the steps of:

(b-1) extracting a face region using motion and color information through inter-frame difference of the face image; and

(b-2) examining edge components and binary information of the image in the face region.

15. The method as set forth in claim 13, wherein the step (c) includes the steps of:

(c-1) extracting eye position from the eye region;

(c-2) normalizing the face image to a predetermined size using eye position; and

(c-3) preprocessing the normalized face image and calculating feature values corresponding to whole sets of features to be used to select features from the face image.

16. The method as set forth in claim 15, wherein the preprocessing at step (c-3) is performed through histogram equalization of the face image.

17. The method as set forth in claim 13, wherein the step (d) includes the steps of:

(d-1) measuring fitness value of the set of features representing each chromosome; and

(d-2) establishing a set of features with improved fitness value by a chromosome evolution process using GA of crossover and mutation as an input vector of the SVM.

18. The method as set forth in claim 13, wherein the feature values are averages of pixels within each window, the windows being overlapped in predetermined sizes to scan each face image.

19. The method as set forth in claim 13, wherein the feature values are averages of edge components of pixels within each window, the windows being overlapped in predetermined sizes to scan each face image.

20. The method as set forth in claim 13, wherein the feature values are entropy values of pixels within each window, the windows being overlapped in predetermined sizes to scan each face image.

21. A method for face authentication by a face authentication system including an image input unit, face and eye region extraction unit, personal information input unit, control unit and SVM classifier, the method comprising the steps of:

(a′

) inputting face image of a specific user upon receiving a request for face authentication by the user including extracting face and eye regions from the inputted face image, wherein the eye regions are extracted from the face region through histogram analysis using edge components and binary information;

(b′

) inputting personal information of the specific user registered in the authentication system upon receiving a request for face authentication;

(c′

) reading selected readily distinguishable feature set, weights and support vectors for the user stored for face authentication matching the personal identification information of the user;

(d′

) extracting the feature values from the inputted face image corresponding to the readily distinguishable feature set stored at training step;

(e′

) using the support vectors and weights, thereby forming a SVM; and

(f′

) using an input vector of the feature values extracted from the inputted face image, thereby performing face authentication using the SVM.

22. The method as set forth in claim 21, wherein the step (a′

) further includes the steps of;

(a′

-1) normalizing the face image to a predetermined size using eye position; and

(a′

-2) preprocessing the normalized face image through histogram equalization, thereby inputting the face image.

23. The method as set forth in claim 21, wherein the method further comprises the step of:

(g′

) displaying a result of face authentication to show the result of face authentication to the user.

24. A method for face authentication by a face authentication system including an ID (identification) card having a SVM, ID recognition unit, an image input unit, face and eye region extraction unit and control unit, the method comprising the steps of:

(a″

) inputting face image of a specific user when the system is requested for face authentication by the user including extracting face and eye regions from the inputted face image, wherein the eye regions are extracted from the face region through histogram analysis using edge components and binary information;

(b″

) forming a SVM for face authentication using support vectors and weights stored in the ID card entered by the user requesting for face authentication; and

(c″

) extracting feature values from the face image of the user corresponding to a set of readily distinguishable features stored in the ID card at a face training step; and

(d″

) using the extracted feature values as an input vector, thereby performing face authentication using the SVM.