AVS快速立体视频编码新方法

提出一种基于AVS(audio video coding standard)的快速立体视频编码方法，对左路参考图像使用AVS编码器编码，对右路目标图像同时在时间域和空间域进行预测. 使用两级神经分类器来快速确定预测模式，第1 级分类产生对每一个大块的不同分割块，第2 级分类在运动估计或视差估计中选出最合适的预测源，并设计了改进的SAD(sum-of-absolute-differences)算法用以计算视差. 实验表明，该文提出的方法能节省80%的时间，引起的质量下降可以忽略，而占用的编码字节数仅多出7%左右.
关键词: 立体视频编码；神经分类器；视差估计；音视频编码标准

Abstract: A new method for fast stereo video encoding based on AVS is presented. In the system, the left-view sequence is encoded by an AVS encoder. For the right-view sequence, combined predictions from the temporal or disparity domains are exploited, with a hierarchical two-stage neural classifier for fast mode decision. The first stage produces different candidates for variable block partitions of each macro-block (MB). The second stage chooses the most suitable prediction sources between forward motion and disparity frames. An improved sum-of-absolutedifferences (SAD) algorithm is proposed in disparity estimation. Experimental results show that the proposed method can achieve time savings up to 80% with negligible quality degradation, and bit-rate increase of only 7%.
Keywords: stereo video coding, neural classifier, disparity estimation, AVS