Three-dimensional (3D) shape modeling based on two-dimensional (2D) warping

1. An electronic device, comprising:

circuitry configured to:

acquire a color image of a face of a user;

acquire depth information corresponding to the color image of the face;

acquire a point cloud of the face;

acquire a three-dimensional (3D) mean-shape model of a reference 3D face;

align the acquired 3D mean-shape model by a rigid alignment of the acquired 3D mean-shape model with the acquired point cloud;

generate a two-dimensional (2D) projection of the aligned 3D mean-shape model, wherein the generated 2D projection comprises a set of landmark points associated with the aligned 3D mean-shape model;

determine, for each pixel of a plurality of pixels of the 2D projection, an index of a triangle on the aligned 3D mean-shape model to which a corresponding pixel of the plurality of pixels of the 2D projection belongs;

record the determined index for each pixel of the plurality of pixels of the 2D projection to generate a triangle index map;

warp the generated 2D projection such that the set of landmark points in the generated 2D projection is aligned with a corresponding set of feature points in the acquired color image of the face;

update a plurality of indices of a plurality of triangles in the triangle index map based on a warp of the warped 2D projection, wherein the plurality of indices of the plurality of triangles includes the determined index of the triangle;

warp the triangle index map based on the update of the plurality of indices of the plurality of triangles;

determine a plurality of 3D positions corresponding to the plurality of pixels of the warped 2D projection based on the acquired depth information and 2D coordinate information of a plurality of corresponding pixels in the acquired color image;

determine, from among the determined plurality of 3D positions, a first 3D position to be closest to a vertex of a first triangle of the plurality of triangles indexed in the warped triangle index map, wherein the first 3D position is determined to be closest based on the warped triangle index map;

assign the determined first 3D position as a first correspondence for the vertex of the first triangle; and

determine, for a non-rigid alignment of the aligned 3D mean-shape model, a 3D correspondence between the aligned 3D mean-shape model and the acquired point cloud, based on the assignment of the first correspondence for each vertex of the plurality of triangles of the aligned 3D mean-shape model.

2. The electronic device according to claim 1 , wherein the circuitry is further configured to acquire the point cloud based on the acquired color image and the acquired depth information.

3. The electronic device according to claim 1 , wherein

the circuitry is further configured to project the aligned 3D mean-shape model onto a 2D image plane to generate the 2D projection, and

wherein the projection is based on at least one alignment parameter associated with an imaging device which captured the color image.

4. The electronic device according to claim 3 , wherein the at least one alignment parameter includes at least one of a focal length, an aperture, an image sensor format, a principal point, a scale factor, or a lens distortion of the imaging device.

5. The electronic device according to claim 1 , wherein

the circuitry is further configured to:

acquire a plurality of landmark points on the aligned 3D mean-shape model; and

execute a landmark marching method on a subset of landmark points of the plurality of landmark points model to select the set of landmark points from among the acquired plurality of landmark points, wherein the subset of landmark points is around a contour of the aligned 3D mean-shape model.

6. The electronic device according to claim 1 , wherein the circuitry is further configured to update the plurality of indices of the plurality of triangles in the triangle index map based on an application of a Moving Least Squares (MLS) morphing method on the triangle index map.

7. A method, comprising:

in an electronic device:

acquiring a color image of a face of a user;

acquiring depth information corresponding to the color image of the face;

acquiring a point cloud of the face;

acquiring a three-dimensional (3D) mean-shape model of a reference 3D face;

aligning the acquired 3D mean-shape model by a rigid alignment of the acquired 3D mean-shape model with the acquired point cloud;

generating a two-dimensional (2D) projection of the aligned 3D mean-shape model, wherein the generated 2D projection comprises a set of landmark points associated with the aligned 3D mean-shape model;

determining, for each pixel of a plurality of pixels of the 2D projection, an index of a triangle on the aligned 3D mean-shape model to which a corresponding pixel of the plurality of pixels of the 2D projection belongs;

recording the determined index for each pixel of the plurality of pixels of the 2D projection to generate a triangle index map;

warping the generated 2D projection such that the set of landmark points in the generated 2D projection is aligned with a corresponding set of feature points in the acquired color image of the face;

updating a plurality of indices of a plurality of triangles in the triangle index map based on a warp of the warped 2D projection, wherein the plurality of indices of the plurality of triangles includes the determined index of the triangle;

warping the triangle index map based on the updating of the plurality of indices of the plurality of triangles;

determining a plurality of 3D positions corresponding to the plurality of pixels of the warped 2D projection based on the acquired depth information and 2D coordinate information of a plurality of corresponding pixels in the acquired color image;

determining, from among the determined plurality of 3D positions, a first 3D position to be closest to a vertex of a first triangle of the plurality of triangles indexed in the warped triangle index map, wherein the first 3D position is determined to be closest based on the warped triangle index map;

assigning the determined first 3D position as a first correspondence for the vertex of the first triangle; and

determining, for a non-rigid alignment of the aligned 3D mean-shape model, a 3D correspondence between the aligned 3D mean-shape model and the acquired point cloud, based on the assignment of the first correspondence for each vertex of the plurality of triangles of the aligned 3D mean-shape model.

8. The method according to claim 7 , further comprising acquiring the point cloud based on the acquired color image and the acquired depth information.

9. The method according to claim 7 , further comprising projecting the aligned 3D mean-shape model onto a 2D image plane to generate the 2D projection, wherein the projection is based on at least one alignment parameter associated with an imaging device which captured the color image.

10. The method according to claim 7 , further comprising:

acquiring a plurality of landmark points on the aligned 3D mean-shape model; and

executing a landmark marching method on a subset of landmark points of the plurality of landmark points for selecting the set of landmark points from among the acquired plurality of landmark points, wherein the subset of landmark points is around a contour of the aligned 3D mean-shape model.

11. The method according to claim 7 , further comprising updating the plurality of indices of the plurality of triangles in the triangle index map based on an application of a Moving Least Squares (MLS) morphing method on the triangle index map.

12. A non-transitory computer-readable medium having stored thereon, computer-executable instructions that when executed by an electronic device, causes the electronic device to execute operations, the operations comprising:

acquiring a color image of a face of a user;

acquiring depth information corresponding to the color image of the face;

acquiring a point cloud of the face;

acquiring a three-dimensional (3D) mean-shape model of a reference 3D face;

aligning the acquired 3D mean-shape model by a rigid alignment of the acquired 3D mean-shape model with the acquired point cloud;

generating a two-dimensional (2D) projection of the aligned 3D mean-shape model, wherein the generated 2D projection comprises a set of landmark points associated with the aligned 3D mean-shape model;

determining, for each pixel of a plurality of pixels of the 2D projection, an index of a triangle on the aligned 3D mean-shape model to which a corresponding pixel of the plurality of pixels of the 2D projection belongs;

recording the determined index for each pixel of the plurality of pixels of the 2D projection to generate a triangle index map;

warping the generated 2D projection such that the set of landmark points in the generated 2D projection is aligned with a corresponding set of feature points in the acquired color image of the face;

updating a plurality of indices of a plurality of triangles in the triangle index map based on a warp of the warped 2D projection, wherein the plurality of indices of the plurality of triangles includes the determined index of the triangle;

warping the triangle index map based on the updating of the plurality of indices of the plurality of triangles;

determining a plurality of 3D positions corresponding to the plurality of pixels of the warped 2D projection based on the acquired depth information and 2D coordinate information of a plurality of corresponding pixels in the acquired color image;

determining, from among the determined plurality of 3D positions, a first 3D position to be closest to a vertex of a first triangle of the plurality of triangles indexed in the warped triangle index map, wherein the first 3D position is determined to be closest based on the warped triangle index map;

assigning the determined first 3D position as a first correspondence for the vertex of the first triangle; and

determining, for a non-rigid alignment of the aligned 3D mean-shape model, a 3D correspondence between the aligned 3D mean-shape model and the acquired point cloud, based on the assignment of the first correspondence for each vertex of the plurality of triangles of the aligned 3D mean-shape model.