面向监控视频内容自适应的通用视频编码量化参数级联  

Content Adaptive Quantization Parameter Cascading for Surveillance Video Coding Using Versatile Video Coding

作　　者：公衍超 张森森 何书婷 李宝贵 刘颖 林庆帆 GONG Yanchao;ZHANG Sensen;HE Shuting;LI Baogui;LIU Ying;LIM Keng-Pang(School of Communications and Information Engineering,Xi’an University of Posts and Telecommunications,Xi’an 710121,China;Electronic Information Experimental Research Center for Forensic Science of Shaanxi Province,Xi’an 710121,China;International Joint Research Center for Wireless Communication and Information Processing Technology of Shaanxi Province,Xi’an 710121,China;XsecPro Pte.Ltd.,787820,Singapore)

机构地区：[1]西安邮电大学通信与信息工程学院,西安710121 [2]陕西省法庭科学电子信息实验研究中心,西安710121 [3]陕西省无线通信与信息处理技术国际联合研究中心,西安710121 [4]新加坡XsecPro公司,新加坡787820

出　　处：《西安交通大学学报》2023年第12期30-40,共11页Journal of Xi'an Jiaotong University

基　　金：国家自然科学基金资助项目(62277036,61801381);西安邮电大学研究生创新基金资助项目(CXJJZL2022005)。

摘　　要：最新通用视频编码(VVC)标准测试模型VTM推荐的JCTVC-X0038量化参数级联方法未考虑监控视频的内容特性,编码效率低。针对这一问题,提出了一种有效考虑监控视频空时域内容特性的VVC量化参数级联方法(SCA-QPC)。基于VVC混合编码框架原理,理论分析得出时间层及监控视频的空时域内容特性是影响量化参数(QP)选择的关键因素,设计编码实验对理论分析结果进行验证;采用梯度和背景帧差分别评测监控视频的空域和时域特性,将梯度和背景帧差作为反向传播神经网络的输入,构建预测低时延结构时间层对应最优QP的模型;提出适用于VVC标准有效提高监控视频编码效率的量化参数级联方法。与当前最先进的JCTVC-X0038量化参数级联方法相比,对于所有测试序列,SCA-QPC对应的平均BD-rate为-4.26%。研究对于VVC编码框架中量化参数级联和码率控制模块的优化具有指导意义。The JCTVC-X0038 quantization parameter cascading(QPC)method recommended by the latest versatile video coding(VVC)standard test model VTM has low coding efficiency because the surveillance video content characteristics are not considered.For this problem,a VVC QPC method considering the spatio-temporal content characteristics of the surveillance videos,referred to as SCA-QPC,was proposed.First,based on the VVC hybrid coding framework principle,it was found that the temporal layer and spatio-temporal content characteristics of the surveillance videos were the key factors affecting the selection of quantization parameters(QP)by theoretical analysis.And the coding experiments were further carried out to verify the results of the theoretical analysis.Then the gradient and background frame difference were used to measure the spatio-temporal characteristics of the surveillance videos respectively and further used as the inputs to the back propagation neural network to construct a model predicting the optimal QPs corresponding to temporal layers in low delay structure.Finally,the SCA-QPC suitable for VVC standard was proposed to effectively improve the surveillance video coding efficiency.Compared with the state-of-the-art JCTVC-X0038 QPC method,the average BD-rate of SCA-QPC is-4.26%for all test sequences.This study can guide the optimization of QPC and rate control modules in the VVC coding framework.

关 键 词：视频编码 通用视频编码 量化参数级联 监控视频 视频内容特性 

分 类 号：TN911.7[电子电信—通信与信息系统]