基于运动特征的HEVC帧间模式快速判决算法

Fast inter mode decision algorithm for HEVC based on motion features

为降低HEVC帧间预测编码过程的计算复杂度,提出了一种新的基于运动特征的编码单元(Coding Units,CU)及预测单元(Prediction Units,PU)快速判决算法。首先,通过二阶帧差法检测运动剧烈程度,并分析运动特征对CU及PU时域相关性的影响。然后,利用前帧相同位置CU及PU的最优模式预测当前CU的深度范围,并根据运动剧烈程度优化预测范围。最后,根据预测范围跳过和终止部分不必要的CU深度和PU模式,从而加快帧间编码速度。在HM12.0平台上的实验测试表明,该算法在BDPSNR仅降低0.068 d B、BDBR只增加1.530%的情况下,提高了48.389%的编码速度。

To reduce the complexity of HEVC inter prediction coding, a novel fast CU （Coding Units） and PU （Prediction Units） mode decision algorithm based on motion features is proposed. Firstly, to detect movement intensity through second order difference method, and to analyze the impact of motion features to CU and PU temporal correlation. Secondly, to predict the current CU depth range and PU mode using the best CU and PU mode of the co-located position in the previous frame. Then, to optimize the predicted range according to the motion features. Finally, in order to speed up the inter prediction coding, to skip and termination of some unnecessary CU and PU mode according to the predicted range. Experimental results on HM12.0 show that the proposed algorithm can reduce encoding time for 48.389%, while BDPSNR loss is only 0.068dB and BRBR increment is 1.530%.