应用压缩域JND模型的VVC感知无损编码  

VVC perceptual lossless coding using JND model in compressed domain

作　　者：公衍超 朱家宝 杨楷芳 刘颖 GONG Yanchao;ZHU Jiabao;YANG Kaifang;LIU Ying(School of Communications and Information Engineering,Xi’an University of Posts and Telecommunications,Xi’an 710121,China;School of Computer Science,Shaanxi Normal University,Xi’an 710119,China)

机构地区：[1]西安邮电大学通信与信息工程学院,陕西西安710121 [2]陕西师范大学计算机科学学院,陕西西安710119

出　　处：《华中科技大学学报（自然科学版）》2024年第9期38-44,共7页Journal of Huazhong University of Science and Technology(Natural Science Edition)

基　　金：国家自然科学基金资助项目(62277036,61801381,61801282)。

摘　　要：针对新一代通用视频编码(versatile video coding,VVC)标准感知无损编码效率低的问题,提出了一种应用压缩域恰可感知差值(just noticeable difference,JND)模型的感知无损编码方法.首先基于ITU—RBT 500—13标准设计图像主观观测实验,获取考虑人类个体差异性及时间影响的VVC压缩图像感知质量数据;然后对数据进行筛选及正态性检验,并分析获得影响图像感知质量的可解释性特征;最终将特征模型量化并作为反向传播神经网络的输入,构建有效反映量化参数与JND关系的压缩域JND模型,并提出显著提升图像编码效率的VVC感知无损编码方法.与VVC标准测试模型方法相比,提出的方法在获得同等感知质量重建图像前提下最高节省93.28%码率.To solve the problem of low perceptual lossless coding efficiency in the new generation versatile video coding(VVC)standard,a perceptual lossless coding method using the compressed domain just noticeable difference(JND) was proposed.Firstly,an image subjective perception experiment was designed based on the ITU—RBT 500—13 standard to obtain the perceptual quality data of VVC compressed images which considered the influence of human individual differences and time.Then,the data was filtered and normality test,and the filtered data was analyzed to obtain the interpretable characteristics that affected the perceptual quality of images.Finally,the characteristics were modeled and used as the input of the back propagation neural network to build a compressed domain JND model that effectively reflected the relationship between quantization parameter and JND.Then,a VVC perceptual lossless coding method was proposed to greatly improve the image coding efficiency.Compared with the method of VVC test model,the proposed method saves 93.28% of the maximum rate under the premise of obtaining the same perceptual quality reconstructed images.

关 键 词：视频编码 通用视频编码 感知无损编码 恰可感知差值 压缩域 

分 类 号：TN919.81[电子电信—通信与信息系统]