Multi-view video coding based on view prediction

Abstract

Multiple view video coding (MVC) is key technology that serves a wide variety of applications, including FTV (free-viewpoint television), 3DTV (three dimensional television) and surveillance. Multi-view video consists of multiple views of the same scene in which there is a high degree of correlation between inter views and intra views. In this paper, a multi-view video coding method based on a new view prediction structure is proposed which makes full use of the spatial redundancy in addition to exploiting the temporal redundancy by performing spatial prediction across different views. A multi-reference disparity prediction method based on variable size block matching is used to get better disparity compensation effect in the proposed structure. Besides, by omitting transmission some B frames, the coding bit-rate is further reduced. Experimental results show that compared to general GoGOP algorithm, the proposed multi-reference prediction algorithm can reconstruct the target images with higher quality, and the MVC scheme based on the new view prediction structure can obtain a coding gain with about 1.4 dB.