Depth coding based on compressed sensing with optimized measurement and quantization

Efficient coding of depth map is an essential part of 3-D video processing system due to the fact that the quality of each synthesized virtual view highly depends on the depth map. In this paper, we propose a compressed sensing (CS) based depth coding scheme. At the encoder side, the depth map is first pair-wisely measured in its Fourier domain, and quantization with a carefully-designed dead-zone is applied on all CS measurements (after considering the distribution of measurement values). An optimized trade-off between the measurement rate and quantization is employed to achieve the best-possible rate-distortion performance. At the decoder side, considering that the depth map usually consists of piece-wise constant areas and sharp edges, we solve a total variation (TV) minimization with constraints being put forward to preserve discontinuities at boundaries and at the same time enforce smoothness within the depth map. Experimental results show that our scheme achieves a significant improvement in rate distortion performance and a better synthesis quality as compared to the standard JPEG scheme.