基于视差稠密化的视图合成  

Generation of intermediate view based on clensification disparity map

作　　者：张倩[1] 安平[1,2] 张兆杨[1,2] 吕朝辉[3] 

机构地区：[1]上海大学通信与信息工程学院,上海200072 [2]上海大学新型显示技术及应用集成教育部重点实验室,上海20072 [3]中国传媒大学信息工程学院,北京100024

出　　处：《高技术通讯》2010年第8期828-832,共5页Chinese High Technology Letters

基　　金：国家自然科学基金(60832003;60672052);上海市曙光计划(06SG43);上海市教委重点项目(09ZZ90);教育部重点实验室开放性课题(07k04)资助项目

摘　　要：针对现有三维视频处理系统需要有较高的传输带宽才能合成高质量的中间视图像的情况,提出了一种新颖的解码端中间视合成方法,该方法仅利用解码端获得的稀疏视差场来合成高质量的中间视图像,这有助于减少传输带宽。该方法以解码端获得的稀疏视差场作为初始值,通过均值偏移算法和总体最小二乘法对其作后处理,获得解码端的稠密视差场,从而合成解码端高质量的中间视画面。实验结果证明,所提方法能仅利用稀疏视差场在解码端获得到平滑稠密视差场和高质量的中间视图。In view of the fact that the existing 3D video processing systems can not synthesize higher quality intermediate view images unless they have a larger bandwidth, this paper proposes a new algorithm for synthesis of intermediate view images at the decoder of the systems. It can produce a synthesized intermediate view image with high quality only by a sparse disparity, which means only a small bandwidth is necessary. The algorithm uses the sparse disparity generated by the decoder as the initial value, and employs the meanshift algorithm and the total least squares （TLS） technique to post-proeess it to obtain a dense disparity at the decoder, thus synthesizing an intermediate view with higher quality from a sparse disparity. The experimental results verified the validity of the algorithm.

关 键 词：视差 中间视 图像分割 分割误差 均值偏移算法 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]