采用SJND模型的动态点云感知编码方法  

Dynamic point cloud perceptual coding method with SJND model

作　　者：刘威 郁梅[1] 蒋志迪[1] 徐海勇[1] LIU Wei;YU Mei;JIANG Zhidi;XU Haiyong(Faculty of Information Science and Engineering,Ningbo University,Ningbo Zhejiang 315211,China)

机构地区：[1]宁波大学信息科学与工程学院,浙江宁波315211

出　　处：《激光杂志》2023年第4期107-113,共7页Laser Journal

基　　金：国家自然科学基金(No.62171243)。

摘　　要：动态点云能有效描述自然场景与3D对象,提供沉浸式视觉体验;但其数据量庞大。需对其进行有效压缩。提出了采用显著性引导的恰可察觉失真(Saliency-guided Just Noticeable Distortion,SJND)模型的动态点云感知编码方法。针对纹理图感知冗余,构建了基于离散余弦变换域的SJND模型,应用于纹理图编码过程中的DCT系数抑制;考虑到相同失真等级下显著区域的几何失真更易被察觉,提出使用投影显著图将几何图进行分层;最后,为不同层级的编码树单元进行自适应量化参数选择和编码。与V-PCC标准方法相比,在保证动态点云视觉质量的前提下,所提出方法提升了动态点云的编码效率。Dynamic point cloud can effectively describe natural scenes and 3D objects,providing an immersive visual experience.However,it has a huge amount of data which needs to be compressed effectively.In this paper,a Saliency-guided Just Noticeable Distortion(SJND)model is defined and used to design a dynamic point cloud perceptual coding method.Considering the perceptual redundancy of texture map,the SJND model is constructed in Discrete Cosine Transform(DCT)domain and used to suppress the DCT coefficients in texture map coding.Considering that the geometric distortion of saliency regions is more easily perceived at the same level of distortion,a projected saliency map is proposed to layer the geometric images.Finally,adaptive quantization parameters(QPs)are selected for Coding Tree units(CTUs)of different levels.Compared with the V-PCC standard method,the proposed method improves the coding efficiency on the premise of ensuring the visual quality of the dynamic point cloud.

关 键 词：动态点云编码 基于视频的点云压缩 显著性引导恰可察觉失真模型 自适应量化参数选择 

分 类 号：TN919[电子电信—通信与信息系统]