基于有记忆信源模型的视频编码量化算法研究

Research on video encoding quantization algorithm with memory source model

针对视音频优化量化算法研究,本文通过模拟最佳软判决量化特点,引入系数间的相关性,在硬判决量化基础上提出一种有记忆信源模型的量化算法。该模型统计了量化块中每个位置编码比特节省估计量,利用贝叶斯二值判别法计算出可区分量化结果的最佳估计阈值,二值做差得到码率节省余量,利用码率节省余量实现对量化偏移量的动态调节,从而优化量化算法。实验表明,基于本文的有记忆信源模型相较于传统硬判决量化有显著性能提升,BD-PSNR有0.0964d B提升,相当于3.5723%码率节省。本文偏移量模型基于离线建模,实时计算所需额外计算复杂度较小,适合硬件编码器架构设计实现。

For the research on audio and video optimization quantization algorithm,this paper simulated the optimal soft decision quantization,introducing the coefficient correlation and puts forward a new model of memory source quantization algorithm based on hard decision quantization.First,this model collect statistics that the estimate which each position encoding bit rate saved in quantization block,use Bayesian discriminant method to calculate the best estimate threshold which can distinguish quantitative results.Then,these two types of data were subtracted to get the surplus that quantitative rate saves.FinaLly,using the surplus that quantitative rate saves to realize adjusting to the quantification offset dynamically so that optimizing the quantization algorithm.Experiments show that memory source model based on this paper compared to traditional hard decision quantization has significant performance improvements,BD-PSNR has 0.0964dB upgrade,and equivalent of 3.5723％ bit rate savings.The model this paper based on offline modeling,and the real-time calculation of extra computational complexity is relatively small,which is suitable for hardware encoder architecture design.