Face recognition system

• US 7,430,315 B2
• Filed: 06/01/2004
• Issued: 09/30/2008
• Est. Priority Date: 02/13/2004
• Status: Expired due to Fees

• Current Assignee: Honda Motor Company
• Original Assignee: Honda Motor Company
• Inventors: Yang, Ming-Hsuan, Ross, David, Ohashi, Takahiro, Lim, Jongwoo
• Primary Examiner(s): Mehta; Bhavesh M.
• Assistant Examiner(s): CARTER, AARON W
• Application Number:
  US10/858,930
• Publication Number:
  US 20050180627A1
• Time in Patent Office: 1,582 Days
• Field of Search: 382/159, 382/279, 382/264
• US Class Current: 382/159
• CPC Class Codes: G06F 18/2411 based on the proximity to a decision surface, e.g. support vector machines
• CPC Class Codes: G06V 30/2504 Coarse or fine approaches, e.g. resolution of ambiguities or multiscale approaches
• CPC Class Codes: G06V 40/161 Detection; Localisation; Normalisation

Abstract

The face detection system and method attempts classification of a test image before performing all of the kernel evaluations. Many subimages are not faces and should be relatively easy to identify as such. Thus, the SVM classifier try to discard non-face images using as few kernel evaluations as possible using a cascade SVM classification. In the first stage, a score is computed for the first two support vectors, and the score is compared to a threshold. If the score is below the threshold value, the subimage is classified as not a face. If the score is above the threshold value, the cascade SVM classification function continues to apply more complicated decision rules, each time doubling the number of kernel evaluations, classifying the image as a non-face (and thus terminating the process) as soon as the test image fails to satisfy one of the decision rules. Finally, if the subimage has satisfied all intermediary decision rules, and has now reached the point at which all support vectors must be considered, the original decision function is applied. Satisfying this final rule, and all intermediary rules, is the only way for a test image to garner a positive (face) classification.

Claims

1. A method for using Support Vector Machine comprising:

receiving training data for a cascade SVM-based classifier, wherein the training data comprises;

a first plurality of support vectors and a first threshold value for the first plurality of support vectors; and

a second plurality of support vectors and a second threshold for the second plurality of support vectors, wherein the second plurality of support vectors includes the first plurality of support vectors and at least one additional support vector;

receiving an input data;

generating a first SVM score for the input data using the first plurality of support vectors;

comparing the first SVM score to the first threshold to determine a first classification of the input data; and

responsive to the first classification of the input data being a positive classification, generating a second SVM score for the input data using the second plurality of support vectors; and

comparing the second SVM score to the second threshold to determine a second classification of the input data.

2. The method of claim 1, farther comprising generating an output data, wherein the input data comprises an image, and wherein the output data indicates whether the input data is a face image or a non-face image based on the second classification.

3. The method of claim 2, wherein responsive to the first classification being a negative classification, determining that the input data is a non-face image.

4. The method of claim 3, further comprising identifying a face in the input data using a face recognition responsive to the output data indicating that the input data is a face image.

5. The method of claim 4, wherein identifying the face comprises projecting the input data to a low dimensional space using a learned project matrix;

determining a cluster center with a minimum distance to the projected input data; and

returning an identity of the cluster center.

6. The method of claim 5, wherein the learned projected matrix is computed using Fisher Linear Discriminant.

7. The method of claim 1, wherein the first plurality of support vectors comprises a first two support vectors, wherein the first SVM score is generated from the first two support vectors.

8. The method of claim 1, wherein the first classification of the input data comprises the positive classification if the first SVM score is greater than the first threshold value.

9. The method of claim 1, wherein the first plurality of support vectors comprises a first four support vectors, wherein the first SVM score is generated from the first four support vectors.

10. The method of claim 1, wherein the first plurality of support vectors comprises a first eight support vectors, wherein the first SVM score is generated from the first eight support vectors.

11. The method of claim 1, wherein the second plurality of support vectors comprises all support vectors, wherein the second SVM score is generated from all the support vectors.

12. The method of claim 1, wherein the first classification comprises a negative classification if the first SVM score is less than the first threshold value.

13. The method of claim 1, wherein the training data is generated by steps including:

receiving a training set, wherein the training set comprises a set of positive examples and a set of negative examples;

generating a plurality of support vectors and a plurality of support vector coefficients from the training set;

selecting a subset of support vectors from the plurality of support vectors;

the subset comprising support vectors capable of rejecting a negative example; and

determining a threshold value for the subset of support vectors.

14. The method of claim 13 further comprising discarding support vectors with small coefficients.

15. The method of claim 1, wherein the input data is a digital image.

16. The method of claim 1 wherein the SVM is used for face detection.

17. The method of claim 1, wherein the input data comprises a 19×

19 pixel subimage.

18. The method of claim 1, further comprising applying histogram equalization to the input data.

19. A system for using Support Vector Machine comprising:

training data receiving means for receiving training data for a cascade SVM-based classifier, wherein the training data comprises;

a first plurality of support vectors and a first threshold value for the first plurality of support vectors; and

a second plurality of support vectors and a second threshold for the second plurality of support vectors, wherein the second plurality of support vectors includes the first plurality of support vectors and at least one additional support vector;

input data receiving means for receiving an input data;

first SVM scoring means for generating a SVM score for the input data using the first plurality of support vectors;

first comparing means for comparing the first SVM score to the first threshold to determine a first classification of the input data;

second SVM scoring means for generating a second SVM score for the input data using the second plurality of support vectors responsive to the first classification of the input data being a positive classification; and

second comparing means for comparing the second SVM score to the second threshold to determine a second classification of the input data.

20. The system of claim 19, further comprising output means for generating an output data, wherein the input data comprises an image, and wherein the output data indicates whether the input data comprises a face image or a non-face image based on the second classification.

21. The system of claim 20, wherein the first comparing means further determines that the input data is a non-face image responsive to the first classification being a negative classification.

22. The system of claim 21, farther comprising face recognition means for identifying a face in the input data using a face recognition responsive to the output data indicating that the input data is a face image.

23. The system of claim 22, wherein said face recognition means includes:

projection means for projecting the input data to a low dimensional space using a learned project matrix; and

cluster center means for determining a cluster center with a minimum distance to the projected input data.

24. The system of claim 23, wherein the learned projected matrix is computed using Fisher Linear Discriminant.

25. The system of claim 19, wherein the first plurality of support vectors comprises a first two support vectors, wherein the first SVM score is generated from the first two support vectors.

26. The system of claim 19, wherein the first classification of the input data comprises the positive classification if the first SVM score is greater than the first threshold value.

27. The system of claim 19, wherein the first plurality of support vectors comprises a first four support vectors, wherein the first SVM score is generated from the first four support vectors.

28. The system of claim 19, wherein the first plurality of support vectors comprises a first eight support vectors, wherein the first SVM score is generated from the first eight support vectors.

29. The system of claim 19, wherein the second plurality of support vectors comprises all support vectors, wherein the second SVM score is generated from all the support vectors.

30. The system of claim 19, wherein the first classification comprises a negative classification if the first SVM score is less than the first threshold value.

31. The system of claim 19, wherein the training data is generated by training means comprising:

training set receiving means for receiving a training set, wherein the training set comprises a set of positive examples and a set of negative examples;

training SV means for generating a plurality of support vectors and a plurality of support vector coefficients from the training set;

subset selection means for selecting a subset of support vectors from the plurality of support vectors;

the subset comprising support vectors capable of rejecting a negative example; and

threshold determining means for determining a threshold value for the subset of support vectors.

32. The system of claim 31, further comprising discarding means for discarding support vectors with small coefficients.

33. The system of claim 19, wherein the input data is a digital image.

34. The system of claim 19, wherein the SVM is used for face detection.

35. The system of claim 19, wherein the input data comprises a 19×

19 pixel subimage.

36. The system of claim 19, further comprising histogram equalization means for applying histogram equalization to the input data.