Inferring the gender of a face in an image

• US 8,588,482 B1
• Filed: 10/17/2011
• Issued: 11/19/2013
• Est. Priority Date: 11/02/2007
• Status: Active Grant

• Current Assignee: Google LLC (Alphabet Inc.)
• Original Assignee: Google Inc. (Alphabet Inc.)
• Inventors: Baluja, Shumeet, Jing, Yushi
• Primary Examiner(s): CHAWAN, SHEELA C
• Application Number:
  US13/274,778
• Time in Patent Office: 764 Days
• Field of Search: 382/100, 382/115, 382/118, 382/155, 382/159, 382/181, 382/190, 382/276, 382/224, 382/103, 382/192, 382/274, 382/305, 705/14.4, 705/319, 706/47, 704/232, 704/E15.042, 704/E15.017, 704/231, 345/426, 345/589, 345/420, 358/1.1, 358/7, 358/1.18, 358/538
• US Class Current: 382/118
• CPC Class Codes: G06Q 50/01 Social networking
• CPC Class Codes: G06V 40/16 Human faces, e.g. facial parts, sketches or expressions
• CPC Class Codes: G06V 40/172 Classification, e.g. identification

Abstract

The subject matter of this specification can be embodied in, among other things, a computer-implemented method that includes receiving a plurality of images having human faces. The method further includes generating a data structure having representations of the faces and associations that link the representations based on similarities in appearance between the faces. The method further includes outputting a first gender value for a first representation of a first face that indicates a gender of the first face based on one or more other gender values of one or more other representations of one or more other faces that are linked to the first representation.

Claims

1. A computer-implemented method comprising:

receiving a plurality of images that display visible physical features of people in the images;

generating a data structure having representations of the visible physical features and associations that link the representations based on similarities in appearance between the visible physical features of the people in the images; and

outputting a first classification value for a first representation of a first visible physical feature of a person in an image that indicates a classification for the first visible physical feature of the person in the image, into one or a plurality of available classification groups, based on one or more other corresponding classification values of one or more other representations of one or more other corresponding visible physical features that are linked to the first representation.

2. The method of claim 1, further comprising determining a similarity score between the first visible physical feature and a second visible physical feature having a second classification value to determine a similarity score between the first visible physical feature and the second visible physical feature.

3. The method of claim 2, wherein outputting the first classification value based on the one or more other classification values comprises weighting a contribution of the second classification value based at least on the similarity score.

4. The method of claim 3, further comprising discarding the contribution of the second classification value if the similarity score is lower than a threshold similarity score.

5. The method of claim 3, further comprising discarding the contribution of the second classification value if the similarity score is lower than a predetermined number or percentage of other similarity scores.

6. The method of claim 1, wherein outputting the first classification value based on the one or more other classification values comprises performing a random walk algorithm on the data structure.

7. The method of claim 1, wherein outputting the first classification value based on the one or more other classification values comprises performing an adsorption algorithm on the data structure.

8. The method of claim 7, wherein outputting the first classification value occurs at an iteration of the algorithm selected from a group consisting of an iteration where the first classification value converges, an iteration where the first classification value changes by less than a threshold amount over a previous iteration, and an iteration occurring after a predetermined number of iterations.

9. The method of claim 1, further comprising seeding a subset of the one or more other classification values with values indicative of one of the plurality of available classification groups.

10. The method of claim 9, wherein the seeding is based at least on user input.

11. The method of claim 9, wherein the seeding is based at least on a computer algorithm.

12. The method of claim 1, wherein each of the one or more other visible physical features is associated with a plurality of classification values that each represent a possible group of the plurality of available classification groups.

13. The method of claim 12, further comprising determining a final classification value for each of the one or more other visible physical features based at least on the plurality of available classification groups, for each visible physical feature.

14. The method of claim 12, further comprising normalizing the plurality of classification values.

15. The method of claim 14, wherein the normalizing of the plurality of classification values for a particular visible physical feature is based at least on the plurality of classification values of the particular visible physical feature.

16. The method of claim 14, wherein the normalizing of the plurality of characteristic values for a particular visible physical feature of a person is based at least on the plurality of characteristic values of the one or more other visible physical feature of other people.

17. The method of claim 1, further comprising using a distribution of values that indicate the visible physical features, to generate a confidence value for the first classification value.

18. The method of claim 17, wherein determining the first classification value includes weighting the first classification value based at least on the confidence value.

19. The method of claim 1, further comprising identifying the visible physical features within the images using a recognition algorithm.

20. The method of claim 1, wherein the visible physical feature comprises a face of a person, and the corresponding classification values represent a gender.

21. A computer-implemented method comprising:

receiving a plurality of images that display visible physical features of animals;

generating a graph that has nodes and edges, wherein the nodes represent the visible physical features of people or animals and the edges link the nodes based at least on similarities in appearance between the visible physical features; and

determining a visible characteristic of a first visible physical feature represented by a first node based at least on the visible characteristics of one or more other visible physical features that have nodes linked to the first node.

22. A computer-implemented method comprising:

receiving a plurality of images that have animal faces;

generating a data structure that includes representations of the faces;

linking the representations based at least on similarities in appearance between the faces; and

determining a first value that indicates a visible characteristic of a first face based at least on visible characteristics of one or more other faces that are linked to the representation of the first face.

23. A system comprising:

a face detector that identifies faces within images;

means for generating a data structure configured to associate a first face with one or more second faces based on similarities in appearance between the first face and the one or more second faces;

means for determining a visible characteristic of the first face based at least on the visible characteristics of the one or more second faces that are associated with the first face; and

an interface that outputs a value that indicates the visible characteristic of the first face.