Abstract:

Coefficient-independent dead-zone hard-decision quantization is a typical video coding quantization algorithm with low complexity but relatively low performance. The Dependent Quantization(DQ) algorithm based on dynamic programming is introduced into the new-generation video coding standard VVC. The coding performance is significantly improved, but the computational complexity increases dramatically. The current algorithm has a high dependence among coefficients, a complex search space of coefficient quantization candidates, and low efficiency of the traversal calculation rate distortion cost. For this, a context-based quantization candidate pruning algorithm is proposed, which greatly reduces the dynamic programming search space and solves the problem of the high complexity of full path search in the DQ algorithm. Through the principle of the DQ algorithm and a statistical analysis of the quantization results, it is found that the DQ quantization results are closely related to context variables such as quantization residues, coefficient positions, and neighborhood quantization results. The complex dynamic programming quantization is abstracted into a multi-variable and multi-interval classification problem of residual, location, and neighborhood quantization results. To address the problem of there being different quantification results in the same interval, a pruning method based on threshold comparison is proposed by analyzing the cumulative distribution function of the samples in the same interval, which cuts several "safe" quantization candidates and reduces the search space. The experimental results show that under the configurations of All Intra and Random Access, the proposed fast DQ algorithm has an average loss of 0.19% and 0.34%, respectively, in rate-distortion performance, and an average reduction of 4.31% and 3.36%, respectively, in coding complexity.

Key words: Versatile Video Coding(VVC) standard, dynamic programming, Dependent Quantization(DQ), branch simplify, context adaptive

摘要：

系数独立的死区硬判决量化是典型的视频编码量化算法，复杂度低，但算法性能相对低下。新一代视频编码标准VVC引入基于动态规划的依赖量化(DQ)算法，编码性能显著提升，但计算复杂度急剧增加。由于算法系数间依赖性高，量化候选搜索空间复杂，导致遍历计算率失真代价效率较低。为此，提出一种基于上下文的量化候选剪枝算法，减小动态规划搜索空间，解决量化候选搜索复杂度较高的问题。依据DQ算法原理和量化结果的统计分析，发现DQ量化结果与量化余数、系数位置、邻域量化结果等上下文变量密切相关，将复杂的动态规划量化抽象为余数、位置、邻域量化结果多变量多区间分类问题，针对同一区间内存在的不同量化结果，通过分析同一区间内样本的累积分布函数，提出基于阈值比较的剪枝方法，裁剪部分“安全”的量化候选，减小搜索空间，简化全路径搜索。实验结果表明，快速DQ算法在All Intra和Random Access配置下，率失真性能平均损失分别为0.19%和0.34%，编码复杂度平均降低了4.31%和3.36%。

关键词: 通用视频编码标准, 动态规划, 依赖量化, 分支简化, 上下文自适应