易错网络中面向虚拟视点绘制的宏块模型

A significance model for virtual view synthesis in error prone network

为了降低传输失真对自由视点视频虚拟视点质量的影响,提出了一种同时适用于彩色图及深度图的宏块区分模型。模型主要包括两部分:首先结合立体视频时空域的相关性,在考虑错误扩散的基础上,提出一个递归形式的宏块级传输失真模型;之后分别讨论彩色图失真和深度图失真对虚拟视点质量的影响,进而提出了一个低复杂度的基于绘制失真分析的重要性模型。实验结果表明,与随机丢包相比,本文算法能在不改变丢包率(PLR)的情况下大幅提高虚拟视点的客观质量,在PLR为20%时,平均峰值信噪比(PSNR)最大能提高15.65dB,且主观质量接近零传输失真情况。

In order to relief the virtual view＇s quality reduction induced by the transmission distortion, we propose a macro block level significance model applied to both depth map and texture map. The model contains two main parts： firstly, by considering the temporal and spatial correlation of the compression structure and the distortion diffusion due to lost packets, we propose a recursive transmission distortion model;secondly,we discuss the relationship between synthesis distortion and both depth-error and tex- ture-error induced distortions. A low-complexity significance model based on the analysis of synthesis distortion is proposed. Experimental results show that compared with random packet loss, the peak signal to noise ratio （PSNR） of the virtual view could be increased dramatically（up to 15.65 dB at 20% packet loss rate）, and the subjective quality is close to the undistorted transmission condition.