一种基于互信息量的自适应快速视频编解码方案

An adaptive fast video encoding/decoding scheme based on mutual information

为了对不同运动剧烈程度的视频序列提出一种重构质量高并且快速的自适应编码算法,引入视频序列的互信息量(mutual information,MI)。利用视频序列帧间的互信息量自适应地调整运动向量估计和运动向量多重估计的门限值(absolute error,AE)。通过大量的统计得出了互信息量与高通帧方差之间的关系,利用方差作为反馈进一步调整AE,从而实现了一种自适应快速算法。利用此算法同时达到了控制编码复杂度和重构视频质量的目的。实验数据表明,此种算法编码时间比传统运动补偿时域滤波(motion compensated temporal filtering,MCTF)结构缩短约15%,而重构视频的质量几乎没有下降。

In order to propose a high-reconstructed quality and fast adaptive video-coding algorithm depending on various motion degrees, mutual information （MI） of video sequences was introduced. According to the MI of different video sequences, the threshold （AE） in the continuous inter-layer motion vector prediction and motion vector multi-prediction was adjusted adaptively. Besides, by a large number of statistical data, the relationship between MI and high-pass frame variance was obtained. Furthermore, the threshold AE was adjusted using the variance as feedback. Thus, an adaptive fast algorithm was realized. This algorithm can control the coding computational complexity and reconstructed video sequences quality at the same time. The experimental results show that this algorithm can save the coding time by 15% compared with the traditional motion compensated temporal filtering （MCTF） scheme, and the reconstructed video sequences quality almost does not decline.