Variable block-size compressed sensing for depth map coding

Abstract

Compressed sensing (CS) is the theory and practice of sub-Nyquist sampling of sparse signals of interest. Perfect reconstruction is possible with much fewer than the Nyquist required number of data samples. In this work, we consider a variable block-size CS architecture for fast compression of depth maps for three-dimensional video (3DV) applications. While existing CS-based depth map coding methods encode depth maps with equal block size, the proposed algorithm partitions a depth map into smooth and edge blocks of variable sizes via rate-distortion optimized quad-tree decomposition. CS is then performed on edge blocks, and eight-bit encoding is performed on smooth blocks. At the decoder, high quality depth map reconstruction is achieved by minimizing the spatial total-variation. Experimental results show that at a small extra expense of encoder complexity, the proposed variable block-size compressed sensing has enhanced significantly the rate-distortion performance over existing low-complexity CS-based depth map coding algorithms.