Method and system for robust human gender recognition using facial feature localization

• US 8,027,521 B1
• Filed: 03/25/2008
• Issued: 09/27/2011
• Est. Priority Date: 03/25/2008
• Status: Active Grant

• Current Assignee: Videomining Corporation
• Original Assignee: Videomining Corporation
• Inventors: Moon, Hankyu, Sharma, Rajeev, Jung, Namsoon
• Primary Examiner(s): WERNER, BRIAN P
• Application Number:
  US12/079,276
• Time in Patent Office: 1,281 Days
• Field of Search: None
• US Class Current: 382/118
• CPC Class Codes: G06V 40/172 Classification, e.g. identification

Abstract

The present invention is a method and system to provide a face-based automatic gender recognition system that utilizes localized facial features and hairstyles of humans. Given a human face detected from a face detector, it is accurately localized to facilitate the facial/hair feature detection and localization. Facial features are more finely localized using the geometrically distributed learning machines. Then the position, size, and appearance information of the facial features are extracted. The facial feature localization essentially decouples geometric and appearance information about facial features, so that a more explicit comparison can be made at the recognition stage. The hairstyle features that possess useful gender information are also extracted based on the hair region segmented, using the color discriminant analysis and the estimated geometry of the face. The gender-sensitive feature vector, made up from the extracted facial and hairstyle features, is fed to the gender recognition machines that have been trained using the same kind of gender-sensitive feature vectors of gallery images.

Claims

1. A method for determining the gender of people based on their facial images, comprising the following steps of:

a) training a plurality of first machines for localizing a face and training a plurality of second machines for localizing facial features in the face using first facial images in a facial image database to extract first gender-sensitive features,

b) training a learning machine for determining gender using the first gender-sensitive features,

c) detecting at least a second face from input images of the people captured by at least a means for capturing images, and localizing the second face using the plurality of first machines, and localizing second facial features using the plurality of second machines based on the localized second face, to extract second gender-sensitive features, and

d) determining the gender of the second face based on the second gender-sensitive features using the trained learning machine for determining gender.

2. The method according to claim 1, wherein the method further comprises a step of annotating locations of facial features, hair regions, and gender labels of the first facial images, in a facial image database.

3. The method according to claim 1, wherein the method further comprises a step of annotating the first facial images in the facial image database with the locations of facial feature landmark points, indexed by the facial features and the face, where a plurality of landmark points delimits a facial feature.

4. The method according to claim 1, wherein the method further comprises a step of annotating first facial images in the facial image database with hair region information by marking the hair tone pixels or marking the boundaries of the hair regions.

5. The method according to claim 1, wherein the method further comprises a step of training a plurality of the first learning machines so that each machine has inherent facial pose, including horizontal and vertical shifts, scale, and orientation, and is tuned to facial images having the inherent facial pose.

6. The method according to claim 1, wherein the method further comprises a step of feeding a facial image to the plurality of the first learning machines, and combining the scores from the learning machines to localize the facial image.

7. The method according to claim 1, wherein the method further comprises a step of training a plurality of second learning machines wherein a subset of the plurality of the second learning machines take part in localizing a specific facial feature, and each learning machine in the subset has inherent pose, including horizontal and vertical shifts, scale, and orientation, of the facial feature, and is tuned to facial feature images having the inherent pose.

8. The method according to claim 1, wherein the method further comprises a step of feeding a facial feature image to the plurality of the second learning machines, and combining the scores from the learning machines to localize the facial feature.

9. The method according to claim 1, wherein the method further comprises a step of constructing first gender-sensitive features by extracting geometric features, appearance features, and hairstyle features of the first facial images, which are localized using the plurality of first learning machines and the plurality of second learning machines.

10. The method according to claim 1, wherein the method further comprises a step of determining a skin-hair tone discriminant axis, which is a direction in color space that best separates the skin tone and hair tone.

11. The method according to claim 10, wherein the method further comprises a step of extracting the hairstyle features of the first facial images, based on the segmented hair region, where the hair region is determined by selecting pixels from the first facial images which belong to the hair tone region in a two-dimensional space of pixels, where the first dimension represents the distance of the pixel from the face center and the second dimension is the skin-hair tone discriminant axis.

12. The method according to claim 11, wherein the method further comprises a step of building a first histogram of pixels from skin tone pixels and a second histogram of pixels from hair tone pixels, and the skin tone region is estimated based on the probability density function derived from the first histogram, and the hair tone region is estimated based on the probability density function derived from the second histogram.

13. The method according to claim 1, wherein the method further comprises a step of extracting geometric features and appearance features of the facial image, where the geometric features are computed from the positions and the sizes of localized facial features and the appearance features are the pixel values from the localized facial features.

14. The method according to claim 13, wherein the method further comprises a step of computing geometric features of the facial features, by using the responses from second machines as the geometric features.

15. An apparatus for determining the gender of people based on their facial images, comprising:

a) means for training a plurality of first machines for localizing a face and training a plurality of second machines for localizing facial features in the face using first facial images in a facial image database to extract first gender-sensitive features,

b) means for training a learning machine for determining gender using the first gender-sensitive features,

c) means for detecting at least a second face from input images of the people captured by at least a means for capturing images, and localizing the second face using the plurality of first machines, and localizing second facial features using the plurality of second machines based on the localized second face, to extract second gender-sensitive features, and

d) means for determining the gender of the second face based on the second gender-sensitive features using the trained learning machine for determining gender.

16. The apparatus according to claim 15, wherein the apparatus further comprises means for annotating locations of facial features, hair regions, and gender labels of the first facial images, in a facial image database.

17. The apparatus according to claim 15, wherein the apparatus further comprises means for annotating the first facial images in the facial image database with the locations of facial feature landmark points, indexed by the facial features and the face, where a plurality of landmark points delimits a facial feature.

18. The apparatus according to claim 15, wherein the apparatus further comprises means for annotating first facial images in the facial image database with hair region information by marking the hair tone pixels or marking the boundaries of the hair regions.

19. The apparatus according to claim 15, wherein the apparatus further comprises means for training a plurality of the first learning machines so that each machine has inherent facial pose, including horizontal and vertical shifts, scale, and orientation, and is tuned to facial images having the inherent facial pose.

20. The apparatus according to claim 15, wherein the apparatus further comprises means for feeding a facial image to the plurality of the first learning machines, and combining the scores from the learning machines to localize the facial image.

21. The apparatus according to claim 15, wherein the apparatus further comprises means for training a plurality of second learning machines wherein a subset of the plurality of the second learning machines takes part in localizing a specific facial feature, and each learning machine in the subset has inherent pose, including horizontal and vertical shifts, scale, and orientation, of the facial feature, and is tuned to facial feature images having the inherent pose.

22. The apparatus according to claim 15, wherein the apparatus further comprises means for feeding a facial feature image to the plurality of the second learning machines, and combining the scores from the learning machines to localize the facial feature.

23. The apparatus according to claim 15, wherein the apparatus further comprises means for constructing first gender-sensitive features by extracting geometric features, appearance features, and hairstyle features of the first facial images, which are localized using the plurality of first learning machines and the plurality of second learning machines.

24. The apparatus according to claim 15, wherein the apparatus further comprises means for determining a skin-hair tone discriminant axis, which is a direction in color space that best separates the skin tone and hair tone.

25. The apparatus according to claim 24, wherein the apparatus further comprises means for extracting the hairstyle features of the first facial images, based on the segmented hair region, where the hair region is determined by selecting pixels from the first facial images which belong to the hair tone region in a two-dimensional space of pixels, where the first dimension represents the distance of the pixel from the face center and the second dimension is the skin-hair tone discriminant axis.

26. The apparatus according to claim 25, wherein the apparatus further comprises means for building a first histogram of pixels from skin tone pixels and a second histogram of pixels from hair tone pixels, and the skin tone region is estimated based on the probability density function derived from the first histogram, and the hair tone region is estimated based on the probability density function derived from the second histogram.

27. The apparatus according to claim 15, wherein the apparatus further comprises means for extracting geometric features and appearance features of the facial image, where the geometric features are computed from the positions and the sizes of localized facial features, and the appearance features are the pixel values from the localized facial features.

28. The apparatus according to claim 27, wherein the apparatus further comprises means for computing geometric features of the facial features, by using the responses from second machines as the geometric features.