多视点视频编码中的视频目标提取与视差匹配

在单通道视频目标分割算法的基础上,首先提出一种多视点视频编码中的视频目标提取方案.然后提出一种基于多级块匹配视差估计算法.应用Moravec算子对视差矢量图进行检测,提取特征视差矢量并对其进行精确度检验.若特征视差矢量不够精确,则其所在块将重新进行小范围的块匹配.若块匹配误差仍太大则将其分裂为四个子块重新匹配.与传统块匹配算法相比,新方法能获得较为精确和平滑的视差场,提高重建后的图像质量.

Video Object Extraction and Disparity Estimation in Multi-View Video Coding

Based on algorithm for extracting video objects in one view-channel,a new scheme of extracting video objects in multi-view coding is proposed.A disparity estimation algorithm based on multi-level block matching is proposed.Moravec operator is used to find out characteristic disparity vectors in disparity images for precision confirmation.If a characteristic disparity vector is not precise enough,the block it belongs to will be matched again around the area indicated by the vector.If a block is still not matched precisely,it is split into four equal sub-blocks for block matching.Compared with traditional block-matching based algorithm,more precise and smooth disparity field can be acquired with the proposed algorithm,so that the quality of reconstructed image can be greatly improved.