低复杂度的多视点视频编码宏块模式决策算法

A low complexity macroblock mode decision algorithm for multiview video coding

为了降低多视点视频编码(MVC)的计算复杂度,提出了一种基于全局-局部率失真代价的低复杂度宏块模式决策算法。首先根据宏块候选模式的块尺寸和率失真代价特点,将所有候选模式分为大尺寸模式(Skip/Direct、Inter16×16和Intra16×16)和小尺寸模式(Inter16×8、Inter8×16、Inter8×8、Intra8×8和Intra4×4);接着统计已编码帧中这两类宏块模式的平均率失真代价,并利用这些平均率失真代价计算当前帧宏块模式决策的全局率失真代价;最后利用得到的全局率失真代价,并结合当前宏块已估计模式的局部率失真代价和空间邻近宏块的模式信息提前终止模式决策,具体包括对大尺寸宏块模式进行提前判定以及对小尺寸候选模式进行逐级选择。实验结果表明:与MVC参考代码中的全搜索模式决策算法相比,本文算法在时域预测视点和视点域预测视点上分别节省了74%和82%的平均编码时间,同时保持了良好的编码率失真性能;与现有模式决策快速算法相比,本文算法降低了更多的编码时间,并具有更好的编码率失真性能。

To reduce the huge computational complexity of multiview video coding （MVC）, this work pro- poses a low complexity macroblock （MB） mode decision by using the global-local rate-distortion costs. First,according to block sizes and rate-distortion （RID） cost characteristics of MB candidate modes, the proposed algorithm classifies all candidate modes into the large-size mode （Skip/Direct, Inter16 × 16 and Intra16× 16 ） and small- size mode （Inter16 × 8, Int er8 × 16, Inter8 × 8, Intra8 × 8 and Intra4 × 4）. Then, the average RD costs of classified MB modes in the coded frames are obtained, and these average RD costs are used to calculate the global RD costs of the current frame. Finally,the proposed algorithm a- dopts the global RD costs, RD costs of estimated modes within current MB and mode information of spa tial neighbor MBs to early terminate the mode decision, and it includes the two-stage early decision of large size mode and the progressive selection of small size mode. Experimental results show that as com- pared with the full mode decision in the MVC reference software, the proposed algorithm saves 74% and 82% encoding time on average, maintaining high RD performance for temporal views and inter-view views, respectively. As compared with state-of art fast algorithms, the proposed algorithm achieves more time savings with better rate-distortion performance.