基于多时-空相关的AVS2快速帧间预测算法

Fast inter-frame prediction algorithm for AVS2 based on multiple temporal-spatial correlations

针对新一代视频标准AVS2引进四叉树分割、多参考帧等技术而带来的帧间预测复杂度增加的问题,提出一种基于多时-空相关的快速帧间预测算法。该算法利用上下层相邻编码单元(Coding Unit,CU)和空时域相邻CU在预测模式选择上的相关性,计算当前CU的模式复杂度,根据复杂度为当前CU分配合适的候选预测模式;同时利用相邻预测单元(Prediction Unit,PU)在参考帧选择上的相关性,计算当前PU的参考帧索引,根据索引为当前PU分配合适的候选参考帧。实验表明,该算法在BD-Rate(Bjontegaard delta bit rate)增加1.22%,BD-PSNR(Bjontegaard delta peak signal-to-noise rate)降低0.04 d B的前提下,平均减少47.54%的编码时间。

The complexity of inter-frame prediction is increased in the new generation video standard AVS2 by introducing the technology of quad-tree division, multiple reference frames, etc. To solve the problem, a fasted inter-frame prediction algorithm based on multiple temporal-spatial correlations is proposed. The mode complexity of current coding unit（CU） is calculated by the correlations of adjacent depth and adjacent temporal-spatial CU in terms of the selection of prediction mode, then the appropriate candidates of prediction mode are distributed according to the mode complexity. At the same time, the reference index of current prediction unit（PU） is calculated by the correlations of adjacent PU in terms of the selection of reference frame, then the appropriate candidates of reference frame are distributed according to the reference index. Experimental results show that, the proposed algorithm can reduce the encoding time by 47.54% on average, with a 1.22% increase of Bjontegaard delta bit rate（BD-Rate） and a 0.04dB decrease of Bjontegaard delta peak signal-to-noise rate（BD-PSNR）.