计算机工程与应用 ›› 2016, Vol. 52 ›› Issue (18): 183-187.

• 图形图像处理 • 上一篇    下一篇

基于边缘建模的纹理深度联合快速三维编码

斯晓华,王国中,李国平,赵海武,滕国伟   

  1. 上海大学 通信与信息工程学院,上海 200436
  • 出版日期:2016-09-15 发布日期:2016-09-14

Fast joint texture depth algorithm based on edge-modeling for 3D-HEVC coding

SI Xiaohua, WANG Guozhong, LI Guoping, ZHAO Haiwu, TENG Guowei   

  1. School of Communication and Information Engineering, Shanghai University, Shanghai 200436, China
  • Online:2016-09-15 Published:2016-09-14

摘要: HEVC的三维视频编码扩展部分(3D-HEVC)引入了视间预测和纹理深度联合预测的工具对相关性较强的纹理、深度视点进行高效编码,同时它沿用了HEVC中的四叉树编码结构和率失真优化来选择最优预测模式和块划分模式,这样导致计算量很大。在3D视频中,纹理图像、深度图像中存在大片较为平坦的纹理区域,深度图像中的平坦区域占比更是达到85%,然而对于平坦区域进行计算量极大的率失真过程是冗余的,针对此问题提出一种基于边缘建模的纹理深度联合快速编码算法。该方法对纹理图像以及深度图像进行边缘建模,对其中的编码单元进行平坦性分析,通过边缘模型中的平坦性和方向性对编码过程进行指导,从而有效地避免了一些冗余的CU深度以及块划分模式的遍历。实验结果表明,在随机访问配置(RA)的情况下,该方法平均降低59.0%的编码时间,而仅在合成视点端带来3.8%性能下降。

关键词: 边缘建模, 快速算法, 三维视频, 模式判决

Abstract: 3D-HEVC introduces interview and intercomponent coding tools for efficient coding of dependent views and depth data, meanwhile it utilizes the quadtree-based coding structure and a complex Rate Distortion Optimization(RDO) process to determine the best coding mode and partition size, which is performed in HEVC for both the texture and depth components. Although the encoding efficiency is improved significantly, the computational complexity is high. In fact, it’s unnecessary to adopt the “try all and find the best” method in the flat area for texture and depth views, and the percentage of flat area in the depth even can be 85%. In order to reduce the computational complexity, a fast 3D-HEVC algorithm is proposed based on the edge-modeling and the correlation between texture and depth. The proposed method divides the coding blocks into flat blocks and edge-contained blocks. Then only the first level of the quadtree and simplest PU partition size are tested. Some of these redundant R-D checks are skipped for the edge-contained blocks based on the edge’s direction. The experiments verify that the proposed algorithm can reduce the computational complexity by 59.0% compared with the original 3D-HEVC encoder under RA test case, with only 3.8% BD-rate increase in the process of view synthesis.

Key words: edge modeling, fast algorithms, three-dimensional video, mode decision