基于率失真优化的AVS-M关键参考帧选择算法

在综合分析AVS-M视频编码传输过程中扩散失真和掩盖失真的基础上，从像素级对端到端的视频传输失真情况进行了研究。采用了一种通用的传输失真度估算模型，根据该失真度估计模型，实现了一种无反馈的关键参考帧选择算法。通过评价选择不同参考帧情况下的端到端失真度期望，在编码器率失真框架内选择最合适的参考帧，使有损传输情况下解码端的失真度达到最小。有反馈信道条件下，提出了基于反馈的关键参考帧选择算法，在帧存开销不大的条件下，极大的提高视频传输的抗差错性能。实验表明，采用的失真度估计模型能够较好地适应AVS-M标准，可以较好地估计出端到端的失真情况。实验还显示，基于率失真优化的关键参考帧选择算法，在3-20%丢包率（128kbps）情况下，无反馈和有反馈关键参考帧选择方法较AVS-M标准，PSNR值分别提高了1.9~6.4dB和2.7~10.9dB,较大提高了AVS-M视频差错恢复性能。

Key Reference Picture Selection Algorithms for AVS-M Based on Rate-Distortion Optimization

The end-to-end error propagation distortion on pixel level precision is studied based on the analysis of the propagation distortion and concealment distortion in AVS-M video encoding and transmission. This paper adopts a general transmission distortion estimation model, and a non-feedback key reference picture selection algorithm is realized based on this model. The distortion expectation of decoder is estimated approximately for different reference pictures selection at encoder. Based on rate-distortion optimization rule, the best suitable reference picture is chosen to minimize the decoder distortion in packet loss environment. A reference picture selection algorithm based on the feedback of key pictures is proposed, which can achieve better error-resilient performance. The experimental results illustrate that the introduced expectation model is suitable for AVS-M standard and it can estimate actual distortion approximately. Around about 1.9~6.4 dB Y-PSNR gains can be benefited using the non-feedback key reference picture selection algorithm and about 2.7~10.9 dB Y-PSNR gains can be benefited using the feedback key reference picture selection algorithm compared with traditional prediction algorithm of AVS at 128kbps and with the packet loss rate from 3% to 20%. The experimental results demonstrate that superior performance of error resilient has achieved using the algorithm introduced in this paper. Key words: video coding; error resilience; rate-distortion optimization; end-to-end distortion; reference picture selection