结合率失真代价预测的HEVC快速CU划分

摘要

In order to reduce the high encoding complexity of high efficiency video coding (HEVC), a fast coding unit (CU) partitioning scheme is proposed.Firstly, based on the Lagrange rate-distortion optimization theory and the experimental observation, the linearly relationships of the encoding bits and distortion between adjacent CU depths were achieved.Secondly, the rate-distortion (RD) cost estimation model of adjacent CU depth was proposed.Finally, the RD cost of next CU depth was predicted after encoding the current CU depth according to the RD cost estimation model.The predicted RD cost was used for judging whether the current CU needs to be spitted.Compared with the method in the reference software model of HEVC, i.e., HM12.0, experimental results reveal that the proposed algorithm can save about 33.2% and 38.9% coding complexity on average whereas the average bit-rate increment are 0.2% and 0.6% for low-delay and random-access coding structures, respectively.%针对高效率视频编码(high efficiency video coding,HEVC)标准编码复杂度较高的问题,提出了一种快速编码单元(coding unit,CU)划分方法.首先,结合拉格朗日率失真优化理论及相关实验数据分析得到相邻CU深度对应的失真及码率分别满足线性关系,并利用此关系,建立了率失真代价的预测模型.利用此模型,可以在编码当前CU深度后快速预测得到下一CU深度的率失真代价,并最终通过代价比较,判断是否需要继续进行CU划分.实验结果表明,相比于HEVC测试模型HM12.0,针对低时延与随机接入编码结构,提出的方法的BD-rate分别增加了0.2%与0.6%,同时,编码时间分别减少了33.2%和38.9%.