结构相似度在AVS帧间模式选择中的应用

Application of structural similarity in inter mode selection for AVS

传统的音视频编码标准(AVS)率失真优化(RDO)算法在帧间模式选择中使用偏差平方和(SSD)作为失真度量,不能很好符合人的主观视觉,而最近提出的结构相似度(SSIM)图像质量评价方法更符合人眼视觉系统(HVS)的特性。为了有效地提高编码效率,本文将SSIM引入拉格朗日代价函数,对失真度量的表示进行修正,并在大量实验的基础上建立了拉格朗日参数的经验公式。实验结果表明,与AVS模式选择算法相比,本文算法平均可节约13.22%的比特率;尤其对于静止块较多的序列,在量化参数QP=10时,节省码率达到30%以上。与此同时,重建的图像质量仅下降0.14%,可忽略不计。

The sum of squared differences（SSD） is used as the distortion metric for inter mode decision in traditional audio video coding standard（AVS） rate distortion optimization（RDO）,but it is not consistent with human vision system（HVS） quite well after a lot of investigations.Compared with the other methods,the structural similarity（SSIM） proposed recently for assessing image quality accords with HVS much more well.Therefore,in order to improve coding efficiency,the SSIM is introduced into Lagrangian cost function to correct the expression of distortion metric.Based on experiments,the empirical formula of Lagrangian parameter is established in this paper.Experimental results indicate that the bit rate using the proposed algorithm is averagely reduced by 13.22% than that using the mode selection algorithm of AVS.Especially,when the sequences have massive stationary blocks,the bite rate is saved more than 30% with QP=10.At the same time,the quality of the reconstructed image is only decreased by 0.14% which could be ignored.