基于恰可察觉失真模型的率失真优化算法

Rate distortion optimization algorithm based on Just Noticeable Distortion Model

传统的混合编码框架没有融入视觉感知因素,并不能很好地去除编码过程中的感知冗余。本文在人眼的三大视觉特性的研究基础上,建立JND模型对原始的率失真优化算法进行改进,利用JND值自适应修正量化参数和调节朗格朗日乘数,实现编码比特和失真的感知平衡。实验结果表明:与官方最新测试模型HM16.9相比,在客观失真性能基本不变,甚至有所提升的情况下,所提算法最高可节省7.812 3%的码率,最低也可节省2.468 2%的码率,有效地提高了编码效率。

The traditional hybrid coding framework does not incorporate visual perception factors and does not remove the perceived redundancy in the encoding process. Based on the research of the three visual characteristics of the human eye, this paper establishes the JND model to improve the original rate-distortion optimization algorithm, and uses the JND value to adaptively correct the quantization parameters and adjust the Langeland multiplier to balance the coding bit and distortion. The experimental results show that compared with the latest test model HM16.9, the proposed algorithm can save up to 7.812 3% of the bitrate and save the minimum of 2.468 2% when the objective distortion performance is basically unchanged or even improved. The code rate effectively improves the coding efficiency.