基于空间相关性的复杂度可分级帧间模式选择

Complexity Scalable Inter Modes Decision Algorithm for H.264 Based on Spatial Correlation

本文提出了一种针对H.264的复杂度可分级的帧间模式选择算法,该算法采用一张概率表将当前宏块的7种预测模式与相邻的上侧和左侧宏块的最优模式联系起来。对于每一种相邻宏块模式的组合情况,该表将当前宏块的7种预测模式按照出现概率的高低进行排序,从而可以确定最可能模式、次最可能模式等。在此基础上,调整候选模式数量以匹配终端平台不同的复杂度约束水平,同时又可以最大可能地保证最优模式处在所选择的部分模式子集中。该算法具有4个复杂度级别:40%,54%,70%和82%,实验证明该方法可以灵活有效地降低编码器复杂度,同时编码质量下降很小,不会影响主观效果。

In this paper, the complexity scalability for H. 264 motion estimation was explored. The complexity scalable algorithm proposed uses a probability table to relate the prediction modes of adjacent MBs to that of the current MB. For each combination of adjacent MBs, the probability table includes a list of prediction modes in order of expected occurrence. Based on this table, an ordering of the most probable, next most probable prediction mode etc. is determined. Hence the number of modes used in inter prediction can be adjusted under different complexity constraints while the optimum mode is guaranteed to lie in the chosen modes with great possibility. This proposed algorithm has 4 computation scales： 40%, 54%, 70%, and 82%. The experimental results show that the proposed method can save the encoding time effectively and flexibly with acceptable performance degradation.