Temporal hole filling for depth image based video rendering

1. A computer-implemented method, comprising:

identifying a plurality of bordering pixels delineating a texture hole region in a target image, the target image representing a target view to be rendered at a first time point;

recording a plurality of depth values of the plurality of bordering pixels, each depth value in the plurality of depth values corresponding to a respective bordering pixel in the plurality of bordering pixels;

automatically clustering the plurality of depth values into two depth value clusters with a depth value threshold separating a first depth value cluster of the two depth value clusters from a second depth value cluster of the two depth value clusters, the first depth value cluster comprising background depth values;

selecting a specific estimation direction from among a plurality of candidate estimation directions for a texture hole pixel in the texture hole region;

estimating a depth value of the texture hole pixel by interpolating depth values of two bordering background pixels in the specific estimation direction, the depth values of the two bordering background pixels belonging to the first depth value cluster;

using the estimated depth value to warp the texture hole pixel into a reference view represented by a reference image rendered at a second time point that is different from the first time point, the texture hole pixel as warped into the reference view being used to identify a reference pixel in the reference image;

predicting a pixel value of the texture hole pixel based at least in part on a reference pixel value of the reference pixel from the reference image.

2. The method of claim 1 , wherein the texture hole region is identified based on a texture hole mask that comprises a binary value for each pixel in the image to indicate whether each such pixel is a texture hole pixel.

3. The method of claim 1 , wherein the image comprises a plurality of texture hole regions that include the texture hole region.

4. The method of claim 1 , wherein the image represents a synthesized image from applying depth-image-based rendering to one or more pre-synthesized texture images and one or more corresponding depth images.

5. The method of claim 1 , wherein the plurality of depth values comprises one of: distance-based depth values or disparity-based depth values.

6. The method of claim 1 , wherein the plurality of depth values is automatically clustered into a background depth value cluster and a foreground depth value cluster using one or more of: a centroid-based algorithm, a density-based algorithm, a K-means clustering algorithm, or Jenks natural breaks optimization.

7. The method of claim 1 , wherein the specific estimation direction is selected based on a plurality of differences each of which represents a difference between depth values of a pair of background pixels in a subset of candidate estimation directions in the plurality of candidate estimation directions; and wherein each candidate estimation direction in the subset of candidate estimation directions has two background pixels for the texture hole pixel in the candidate estimation direction.

8. The method of claim 1 , wherein an image metadata portion is encoded in a video stream along with the target image; wherein a downstream decoder that receives the video stream performs at least a part of temporal hole filling operations for the texture hole region of the target image based on the image metadata portion as signaled in the video stream.

9. The method of claim 1 , wherein the method is performed by one of: a video decoder, a video encoder, or a video transcoder.

10. The method of claim 1 , wherein the method is independently performed by a processing thread for each texture hole pixel in the image.

11. A non-transitory computer readable storage medium, storing software instructions, which when executed by one or more processors cause performance of the method recited in claim 1 .

12. A computing device comprising one or more processors and one or more storage media, storing a set of instructions, which when executed by one or more processors cause performance of the method recited in claim 1 .