| Multiview video coding (MVC) has been widely considered as one of the key technologies of three-dimensional (3D) video applications, since it can provide arbitrary views of scenes and has natural and realistic sight. It has been attracting more and more attentions and becomes a significant field of video in recent years. MVC has broad application prospects in 3D video field, such as free-viewpoint television (FTV), 3D television (3DTV) and virtual reality system. However, multiview videos contain a large amount of data as the increased number of cameras, it makes challenge for data storage and bit-stream transmission. Therefore, multiview video data should be compressed efficiently with neglectable degradation on visual quality. To improve coding compression efficiency, MVC exploits the spatio-temporal correlations in the same view and the inter-view correlations among different views. Whereas, high compression efficiency is achieved at the expense of increased computational complexity. Among them, the huge computation is mainly from macroblock mode selection, multiple reference frames application and motion/disparity estimation. In this paper, some fast algorithms are proposed for MVC, including(1) A fast marcoblock mode selection algorithm based on the mean absolute difference of macroblock luminance is proposed. By analyzing the statistical features of the best mode in MVC and the characteristics of marcoblock luminance distribution, then dividing picture into different areas and selecting corresponding modes according to the characteristic of luminance in each frame. Therefore, mode selection can be effective in reducing the number of times, and through experiments that encoding speed of four times.(2) A merged fast macroblock mode selection algorithm for B frames is presented. Firstly, we use temporal correlation of macroblock modes in the corresponding locations among B frames at a view to reduce the complexity of determining the best MB mode. Secondly, the Lagrangian cost of SKIP mode is compared with that of Inter16×16 mode so as to quickly determine whether the mode selection process is halfway stopped or not. Thirdly, exploiting the correlation of the best reference frames among different Inter modes in current MB, the proposed algorithm adopts an efficient method to search the best reference frames of Inter modes quickly. Lastly, a merged algorithm is proposed by combining these three methods, and experimental results show that the proposed algorithm can speed up 3.71~7.22 times compared with the joint multiview video model.(3) A fast disparity estimation algorithm based on the correlations between the neighboring views for MVC is proposed. Firstly, the correlations of various disparity vectors are anailzed in detail. At the same time, the disparity vectors correlation of Inter modes are also analyzed, and the possible prediction disparity vectors are compared. Lastly, fast disparity estimation algorithm based on the original algorithm is presented. Experimental results show that about 40% encoding time can be saved by the proposed algorithm.(4) To improve the precision of block matching in the motion/disparity estimation process, the prediction vectors is adjusted by the temporal correlation of motion/disparity vectors. Different search ranges are used for motion/disparity estimation. Finally, compared with searching methods in JMVM, the improved method has a slight declining in bitrate while keeping the same PSNR. |
PDF Full Text Request