Abstract:

Multi- view depth (MD) is usually coded by using the hierarchical B prediction structure,but the corre-sponding coding complexity is very high.In order to reduce the coding complexity of MD,a fast coding algorithm is proposed.The proposed algorithm takes into consideration the characteristic of coded frames,namely,the coded frames of the higher temporal layer (TL) in the hierarchical B prediction structure are of higher temporal correlation with their reference frames due to the shorter temporal interval between them,which results in a higher percentage of the macro- blocks (MBs) coded by using the large size mode (LSM).Then,by considering the feature that the depth image consists of large flat areas and edges,the simplification strategies of candidate MB modes with different intensities are proposed for the coded frames at different TLs,so as to skip the unnecessary small- size MB modes.Moreover,an early termination algorithm for those MBs adopting LSM is proposed to further reduce the coding com-plexity.Experimental results demonstrate that,as compared with the original coding algorithm,the proposed method reduces the coding complexity by 89.61%,and maintains the rate distortion performance.

Key words: three- dimensional video coding, depth coding, fast coding, hierarchical B prediction structure