Fast Disparity Refinement Algorithm for Client-Oriented Multi-view Video System

Compared to traditional multi-view systems, client-oriented multi-view video system could significantly simplify the complexity at the client since part of disparity estimation is left to the sever. This paper proposes a fast disparity refinement algorithm for the client-oriented multi-view video system. Firstly, the coarse disparity map is generated by adopting 8times8 block based disparity estimation based on smoothness restriction at the server, and the 8times8 block based disparity map is transmitted to the client. Then according to the spatial correlation of disparity vectors and disparity discontinuity at the boundary area of objects, pixel-wise disparity map is obtained with a fast disparity refinement algorithm. Experimental results show that, compared with the traditional block-matching based algorithm at the client, the disparity refinement algorithms spend no more than 2% time to generate a pixel-wise disparity map, moreover, the disparity maps are more smoother within the objects. Compared with the previous refinement algorithm, the proposed algorithm not only improves the quality of the reconstructed right view images, but also saves about 20% time for generating of pixel-wise disparity map.