Siamese reconstruction convolutional neural network for pose-invariant face recognition

1. A computer-implemented method for pose-invariant facial recognition, the method comprising:

generating, by a processor using a recognition neural network, a rich feature embedding for each of one or more images, wherein the rich feature embedding constitutes both identify information and non-identity information;

generating, by the processor using a Siamese reconstruction network, one or more pose-invariant features by employing the rich feature embedding for identity information and non-identity information;

identifying, by the processor, a user by employing the one or more pose-invariant features; and

controlling an operation of a processor-based machine to change a state of the processor-based machine, responsive to the identified user in the one or more images.

2. The computer-implemented method of claim 1 , wherein the one or more images are selected from a group consisting of a still image, a frame from a video file, or a frame from a live video stream.

3. The computer-implemented method of claim 1 , wherein the Siamese reconstruction network disentangles the identity information and the non-identity information in the rich feature embedding.

4. The computer-implemented method of claim 3 , wherein the disentangled rich feature embedding improves discriminality and pose-invariance.

5. The computer-implemented method of claim 1 , wherein the Siamese reconstruction network minimizes errors between a self-reconstruction and a cross-reconstruction.

6. The computer-implemented method of claim 5 , wherein the self-reconstruction combines the identity information with the non-identity information.

7. The computer-implemented method of claim 6 , wherein the cross-reconstruction combines the non-identity information with identity information from another one of the one or more images.

8. The computer-implemented method of claim 1 , wherein the recognition neural network is trained with multiple different sources to learn both the identity information and the non-identity information.

9. The computer-implemented method of claim 1 , wherein the recognition neural network employs a pose-variant face synthesization process.

10. The computer-implemented method of claim 9 , wherein the pose-variant face synthesization process utilizes near-frontal faces to generate non-frontal views.

11. A computer program product for pose-invariant facial recognition, the computer program product comprising a non-transitory computer readable storage medium having program instructions embodied therewith, the program instructions executable by a computer to cause the computer to perform a method comprising:

generating, by a processor using a recognition neural network, a rich feature embedding for each of one or more images, wherein the rich feature embedding constitutes both identify information and non-identity information;

generating, by the processor using a Siamese reconstruction network, one or more pose-invariant features by employing the rich feature embedding for identity information and non-identity information;

identifying, by the processor, a user by employing the one or more pose-invariant features; and

controlling an operation of a processor-based machine to change a state of the processor-based machine, responsive to the identified user in the one or more images.

12. A pose-invariant facial recognition system, comprising:

a processor configured to:

generate, using a recognition neural network, a rich feature embedding for each of one or more images, wherein the rich feature embedding constitutes both identify information and non-identity information;

generate, using a Siamese reconstruction network, one or more pose-invariant features by employing the rich feature embedding for identity information and non-identity information;

identify a user by employing the one or more pose-invariant features; and

control an operation of a processor-based machine to change a state of the processor-based machine, responsive to the identified user in the one or more images.

13. The pose-invariant facial recognition system of claim 12 , wherein the one or more images are selected from a group consisting of a still image, a frame from a video file, or a frame from a live video stream.

14. The pose-invariant facial recognition system of claim 12 , wherein the Siamese reconstruction network disentangles the identity information and the non-identity information in the rich feature embedding.

15. The pose-invariant facial recognition system of claim 12 , wherein the disentangled rich feature embedding improves discriminality and pose-invariance.

16. The pose-invariant facial recognition system of claim 12 , wherein the Siamese reconstruction network minimizes errors between a self-reconstruction and a cross-reconstruction.

17. The pose-invariant facial recognition system of claim 16 , wherein the self-reconstruction combines the identity information with the non-identity information.

18. The pose-invariant facial recognition system of claim 16 , wherein the cross-reconstruction combines the non-identity information with identity information from another one of the one or more images.

19. The pose-invariant facial recognition system of claim 12 , wherein the recognition neural network is trained with multiple different sources to learn both the identity information and the non-identity information.

20. The pose-invariant facial recognition system of claim 12 , wherein the recognition neural network employs a pose-variant face synthesization process.