System and method for calculating an optimization for a facial reconstruction based on photometric and surface consistency

1. A computer-implemented method for generating a three-dimensional model of an object, the method comprising:

implementing, via one or more processors, a stereo reconstruction technique, wherein the stereo reconstruction technique provides an initial estimate of a disparity map;

optimizing disparity values of the disparity map by adjusting the disparity values in directions of determined improved photometric-consistency and in directions of determined improved surface-consistency, wherein the directions of improved surface-consistency are determined using a smoothing term having a second-order anisotropic formulation;

generating a geometry mesh of the object from the optimized disparity map.

2. The method of claim 1 , wherein implementing the stereo reconstruction is based on two images of the object used to generate the geometry mesh.

3. The method of claim 2 , wherein implementing the stereo reconstruction comprises:

rectifying the two images of the object;

applying a high-pass filter to the two images of the object to extract high spatial frequency features; and

performing stereo matching based on the high spatial frequency features.

4. The method of claim 1 , further comprising determining a first set of mesoscopic details associated with the object by applying a filter to an image of the object, wherein mesoscopic details included in the first set of mesoscopic details are detectable in the image of the object and are not reconstructable when generating the geometry mesh of the object.

5. The method of claim 4 , wherein the filter corresponds to a range of spatial frequencies that is associated with the first set of mesoscopic details.

6. The method of claim 4 , further comprising modulating a refinement of the geometry mesh, based on photometric consistency and surface consistency, with the first set of mesoscopic details.

7. The method of claim 6 , wherein modulating the refinement of the geometry mesh with the first set of mesoscopic details is based on a mesoscopic consistency term.

8. The method of claim 4 , wherein the object comprises a human head, and the mesoscopic details comprise spots, freckles, moles, pores, fine wrinkles, or facial hair.

9. A non-transitory computer-readable storage medium storing instructions that, when executed by a processor, cause a computer system to generate a three-dimensional model of an object, by performing the steps comprising:

implementing a stereo reconstruction technique, wherein the stereo reconstruction technique provides an initial estimate of a disparity map;

optimizing disparity values of the disparity map by adjusting the disparity values in directions of determined improved photometric-consistency and in directions of determined improved surface-consistency, wherein the directions of improved surface-consistency are determined using a smoothing term having a second-order anisotropic formulation;

generating a geometry mesh of the object from the optimized disparity map.

10. The computer-readable storage medium of claim 9 , wherein implementing the stereo reconstruction is based on two images of the object used to generate the geometry mesh.

11. The computer-readable storage medium of claim 10 , wherein implementing the stereo reconstruction comprises:

rectifying the two images of the object;

applying a high-pass filter to the two images of the object to extract high spatial frequency features; and

performing stereo matching based on the high spatial frequency features.

12. The computer-readable storage medium of claim 9 , the steps further comprising determining a first set of mesoscopic details associated with the object by applying a filter to an image of the object, wherein mesoscopic details included in the first set of mesoscopic details are detectable in the image of the object and are not reconstructable when generating the geometry mesh of the object.

13. The computer-readable storage medium of claim 12 , wherein the filter corresponds to a range of spatial frequencies that is associated with the first set of mesoscopic details.

14. The computer-readable storage medium of claim 12 , the steps further comprising modulating a refinement of the geometry mesh, based on photometric consistency and surface consistency, with the first set of mesoscopic details.

15. The computer-readable storage medium of claim 14 , wherein modulating the refinement of the geometry mesh with the first set of mesoscopic details is based on a mesoscopic consistency term.

16. The computer-readable storage medium of claim 12 , wherein the object comprises a human head, and the mesoscopic details comprise spots, freckles, moles, pores, fine wrinkles, or facial hair.

17. A system for generating a three-dimensional model of an object, the system comprising:

a processor configured to:

implementing a stereo reconstruction technique, wherein the stereo reconstruction technique provides an initial estimate of a disparity map,

optimizing disparity values of the disparity map by adjusting the disparity values in directions of determined improved photometric-consistency and in directions of determined improved surface-consistency, wherein the directions of improved surface-consistency are determined using a smoothing term having a second-order anisotropic formulation,

generating a geometry mesh of the object from the optimized disparity map.

18. The system of claim 17 , further comprising a memory storing instructions that, when executed by the processor, configure the processor to:

generate the geometry mesh of the object;

calculate an optimization for the geometry mesh based on photometric consistency and surface consistency; and

refine the geometry mesh with the optimization.

19. A computer-implemented method for generating a three-dimensional model of an object, the method comprising:

generating, via one or more processors, a coarse geometry mesh of the object, wherein generating the coarse geometry mesh of the object comprises implementing a stereo reconstruction technique, wherein implementing the stereo reconstruction technique provides an initial estimation of a disparity map;

calculating an optimization for the coarse geometry mesh based on photometric consistency and surface consistency associated with the coarse geometry mesh;

refining the coarse geometry mesh based on the optimization; and

optimizing disparity values associated with the disparity map based on adjusting the disparity values in a determined direction of improved photometric-consistency and in a determined direction of improved surface-consistency.