一种时间一致性立体视频深度图序列估计算法  

Temporal consistency algorithm for depth map sequence estimation of stereoscopic video

作　　者：段峰峰[1,2] 王永滨[1] 杨丽芳[1] 于鹏[1] 

机构地区：[1]中国传媒大学计算机学院,北京100024 [2]湖南师范大学,长沙410081

出　　处：《计算机应用研究》2015年第10期3142-3146,共5页Application Research of Computers

基　　金：国家科技支撑计划资助项目(2012BAH37F02;2012BAH17B02);国家文化部科技创新项目(2014KJCXXM08);国家"863"计划资助项目(2011AA01A107)

摘　　要：针对双目立体视频深度图序列估计中时间不一致性而导致的闪烁问题,提出了一种基于Lucas-Kanade光流加权直方图相似性约束的算法。该算法在使用基于外极线约束的图割方法进行深度估计时,加入带有相干系数的罚函数;计算时间相邻帧颜色直方图加权和,权值由相邻帧像素相似性测度来决定,将计算值与当前帧颜色直方图作相似性比较,根据瞬时运动中当前帧像素与其相邻帧的对应邻域中的像素亮度保持一致性的关系,对比较结果进行最优化求解,确定相干系数,实现对应像素一致性匹配,从而获得时间一致性深度图序列。实验结果表明,该算法能够有效解决闪烁问题,获得较好的时间一致性深度图序列。To eliminate the flickering artifacts caused by temporal inconsistency in binocular stereoscopic video depth map sequence estimation,this paper proposed a similarity constraint algorithm based on Lucas-Kanade optical flow weighted histogram（ LKOFWH）. The algorithm contained the penalty function with coherence factor when used the method of graph cut based on epipolar rectification for depth estimation. It calculated weighted histogram of adjacent frames as the weights determined by pixel similarity measurement. It compared the calculated values with the color histogram of the current frame. According to the principle that pixel intensity should keep consistency between current frame and adjacent frames in instantaneous motion,the comparative values were calculated for optimal solution. It could obtain the coherence factor. Then it could achieve the consistency of corresponding pixel matching and depth map sequence. The experimental results show that the algorithm can effectively solve the problem of flickering artifacts and obtain better temporal consistency depth map sequence.

关 键 词：时间一致性 深度图序列 立体视频 Lucas-Kanade光流加权直方图 

分 类 号：TP391[自动化与计算机技术—计算机应用技术] TP301.6[自动化与计算机技术—计算机科学与技术]