System and method for identifying faces in unconstrained media

1. A method comprising:

determining a three-dimensional (3D) model of a face of an individual based on a plurality of different images of the individual;

extracting two-dimensional (2D) patches from the 3D model; and

generating a plurality of signatures of the face using different combinations of the 2D patches, wherein the plurality of signatures correspond to respective views of the 3D model from different angles;

determining a plurality of attributes of the individual that semantically describe characteristics of the individual; and

indexing the plurality of signatures based on the plurality of attributes.

2. The method of claim 1 , wherein the determining the 3D model comprises:

identifying elements of the 3D model lacking information from the plurality of images; and

providing the information for the identified elements using domain knowledge compiled from individuals having attributes similar to those of the individual.

3. The method of claim 1 , further comprising modifying the 3D model by normalizing lighting variations in the 3D model.

4. The method of claim 1 , further comprising neutralizing a facial expression resulting from the plurality of different images of the individual.

5. The method of claim 1 , further comprising modifying the 3D model based on an age of the individual.

6. The method of claim 1 , further comprising determining respective uncertainty values for the plurality of signatures, wherein the uncertainty values are based on a quality of respective 2D patches included in the plurality of signatures.

7. The method of claim 1 , further comprising determining that a face image matches at least one of the plurality of signatures.

8. The method of claim 7 , wherein the determining that the face image matches comprises modifying a resolution of the plurality of signatures based on a resolution of the face image.

9. The method of claim 7 , wherein the determining that the face image matches comprises matching using a plurality imaging modalities.

10. The method of claim 1 , wherein the plurality of signatures of the face are iteratively refined using a number of additional face images of the individual.

11. The method of claim 10 , wherein the plurality of signatures of the face has a fixed size irrespective of the number of additional face images.

12. The method of claim 1 , further comprising:

determining uncertainty metrics corresponding, respectively, to the plurality of signatures;

associating the plurality of signatures with the corresponding uncertainty metrics.

13. The method of claim 1 , further comprising determining which of the plurality of signatures corresponds to a portion of the face having a greatest number of discriminative features.

14. A facial recognition system comprising:

a processor;

a storage system;

program instructions stored on a computer readable hardware storage device for execution by the processor, the program instructions comprising:

program instructions that determine a three-dimensional (3D) model of a face of an individual based on a plurality of different images of the individual;

program instructions that extract two-dimensional (2D) patches from the 3D model; and

program instructions that generate a plurality of signatures of the face using different combinations of the 2D patches, wherein the plurality of signatures correspond to respective views of the 3D model from different angles;

determining a plurality of attributes of the individual that semantically describe characteristics of the individual; and

indexing the plurality of signatures based on the plurality of attributes.

15. The system of claim 14 , wherein the determining the 3D model comprises:

identifying elements of the 3D model lacking information from the plurality of images; and

providing the information for the identified elements using domain knowledge compiled from individuals having attributes similar to those of the individual.

16. The system of claim 14 , further comprising modifying the 3D model by normalizing lighting variations in the 3D model.

17. The system of claim 14 , further comprising normalizing a facial expression resulting from the plurality of different images of the individual.

18. The system of claim 14 , further comprising modifying the 3D model based on an age of the individual.

19. The system of claim 14 , further comprising determining respective uncertainty values for the plurality of signatures, wherein the uncertainty values are based on a quality of respective 2D patches included in the plurality of signatures.

20. The system of claim 14 , further comprising determining that a face image matches at least one of the plurality of signatures.

21. The system of claim 20 , wherein the determining that the face image matches comprises modifying a resolution of the plurality of signatures based on a resolution of the face image.

22. The system of claim 20 , wherein the determining that the face image matches comprises matching using a plurality imaging modalities.

23. The system of claim 14 , wherein the plurality of signatures of the face are iteratively refined using a number of additional face images of the individual.

24. The system of claim 14 , wherein the plurality of signatures of the face has a fixed size irrespective of the number of additional face images.

25. The system of claim 14 , further comprising:

determining uncertainty metrics corresponding, respectively, to the plurality of signatures;

associating the plurality of signatures with the corresponding uncertainty metrics.

26. The system of claim 14 , further comprising determining which of the plurality of signatures corresponds to a portion of the face having a greatest number of discriminative features.