Low-Complexity Region-of-Interest Extraction for Multiview Video Coding

Region-of-interest (ROI) in multiview video (MW) is different from that of conventional single view video because MW provides three-dimensional perception and makes people more interested in depth discontinuity and pop-out regions. In this paper, we define a novel depth perceptual ROI for MW and discuss four ROI extraction schemes according to temporal and inter-view correlation of the MW. Then, depth based ROI extraction algorithm is proposed by jointly using depth, motion and texture information of MW plus depth data. In order to reduce the computational complexity and improve ROI extraction efficiency for group of pictures, temporal tracking method is utilized. Furthermore, we also present a novel interview tracking method, in which geometry correlation between views and extracted ROI of neighboring views are utilized to facilitate ROI extraction in view dimension. ROI extraction results for MW show that the proposed ROI tracking and extraction algorithms maintain high extraction accuracy and low complexity.