Segment-wise Coding of Texture and Depth Components in 3D Video

In this work, two elements of technology are developed in the context of 3D-HEVC.

Firstly, a segment-wise representation of depth maps by constant depth values and secondly, a segment-based motion compensation scheme, which allows for more precise representation of motion or disparity discontinuities between two segments.

The first algorithm works as an alternative to the conventional residual coding scheme of HEVC. It partitions a block into up to two arbitrarily shaped segments and codes a single DC offset value per resulting segment. Moreover, a depth lookup table is used for the transformation of residual values into index values with a significantly lower dynamic range, which can provide further compression benefit. The second algorithm allows approximating arbitrarily shaped discontinuities in the disparity or motion vector fields used for video texture prediction by incorporating segmentation information from the collocated depth map.

Whether a simplified representation of depth information might also be sufficient for a good 3D video rendering quality is additionally investigated in a subjective study on the necessary precision of depth information. It turns out that this only holds as long as the simplification process preserves the location of depth discontinuities. Based on these observations, it can be concluded that compression efficiency improvements with respect to perceptually optimized visual quality are even more significant.