基于视差和变化检测的立体视频对象分割  

Stereo Video Object Segmentation Based on Disparity and Change Detection

作　　者：高欣[1] 安平[1] 刘佳[1] 张兆扬[1] 

机构地区：[1]上海大学通信与信息工程学院,上海200072

出　　处：《上海大学学报（自然科学版）》2006年第2期116-119,124,共5页Journal of Shanghai University:Natural Science Edition

基　　金：上海市科委自然基金资助项目(04ZR14056);上海市教委发展基金资助项目(04AB56);上海市重点学科建设项目(T0102)

摘　　要：立体视频对象分割是新一代视频编码、视频检索、互联网多媒体交互等新兴领域的关键技术.该文提出了一种融合多种信息的立体视频对象分割算法,首先采用基于边缘特征约束的视差估计方法,获得了可靠的视差场,再利用视差信息对视频对象进行了初始分割,在此基础上结合帧间变化检测的信息对视差分割结果进行修正.实验结果表明,本算法对有重叠的多对象具有良好的分割效果,得到了较精确的视频对象.Stereo video object segmentation is a key technology in the new generation video coding, video browsing, and Internet multimedia interacting, etc. In this paper, a stereo video object segmentation algorithm that combines various sources of information is proposed. Estimation of disparity based on edge character restriction is ilrst adopted to obtain reliable disparity scene. segmentation. Change detection information is then integrated Experimental results show effectiveness of the algorithm in the The disparity information is used to initialize to modify boundaries of disparity segmentation. case of multiple object overlapping.

关 键 词：立体视频 对象分割 视差 变化检测 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]