改进的基于对极几何的DMVC空间边信息重建

An Improved Reconstruction Algorithm of DMVC Spatial Information Based on Epipolar Geometry

分布式多视点视频系统中,边信息分为时间边信息和空间边信息,其中,基于对极几何的空间边信息生成算法是目前最有效的方法之一。在基于对极几何的边信息估计过程中,为了更准确快速地确定对极线上的搜索起点,论文通过统计分析相邻视点同一时刻帧内特征点偏移的局部集中性,提出了一种基于空间特征点分布特性的搜索起点确定算法。首先对关键帧图像进行分块,计算每个分块的特征点偏移均值;然后基于空间坐标点深度时域局部连续性,针对当前WZ帧分块,直接利用处于相同位置的关键帧分块的平均偏移值计算初始偏移中心,最后实现空间边信息的重建。实验结果表明,在不增加计算复杂度的前提下,相对于基于中值预测投影点的算法系统总体性能约有0.25～0.8dB的提升。

In Distributed Multi-view Video Coding（DMVC） system,side information can be generated based on spatial adjacent frames and temporal adjacent frames.Epipolar based algorithm is one of the most effective way to generate spatial side information.In order to acquire better disparity search center,the local disparity distribution of each matched feature points pair is analyzed,and new algorithm based on spatial distribution of feature point disparity is proposed to get disparity search center.The key frames are firstly divided into blocks to grab the local disparity distribution property.And then the average disparity of match point in current block is calculated.Considering the temporal continuity of motion,the average disparity of the block located at the same position in adjacent key frames is directly used as the initial disparity of current WZ block.Experimental results shows 0.25~0.8dB improvement over existing algorithm based on mean disparity estimation.