Parameterized estimation of common motion for image and depth sequences

Abstract

Representation of 3D video using reference image and depth map sequences (2D+depth) has become standard. Although standards recommend separate encoding of image and depth, exploitation of their correlation opens up opportunities, especially for low-bit-rate applications. To this end, a common motion vector is sometimes used to compensate both the sequences. In this context, we propose a parametric method for estimation of common motion that improves upon the current image-based approach. Specifically, we identify statistical redundancy between image and depth sequences, demonstrate how the said redundancy cannot fully be exploited by a technique based on image sequences alone, and then propose a method to remove such redundancy in an optimized manner. Finally, we demonstrate the efficacy of our technique using well-known stereoscopic video sequences.