Depth based region of interest extraction for multi-view video coding

In this paper, a framework of region-of-interest (ROI) based multi-view video coding has been proposed to improve compression efficiency by properly segmenting the multi-view videos into different macroblock level ROI and encoding them separately. We define depth based ROI for multi-view video. Then, we propose a semantic low-complexity rode extraction algorithm based on multi-view video plus depth. ROI quality can be improved by allocating more bits on ROI regions while fewer bits are allocated to non-ROI regions for bit rate saving. Experimental results on multi-view video coding show that 11.34% and 21.41% bit rate can be saved for Ballet and Breakdancers sequences while maintaining high objective quality of ROI regions and high subjective quality of the multi-view videos.