DMVC中基于图像配准的空域边信息重建算法

Spatial Side Information Reconstruction Algorithm Based on Image Registration of DMVC System

为了更好地压缩分布式多视点视频编码系统空间冗余度,提出了一种利用视点间存在的固有的对极几何关系计算空域边信息的方法。在此方法中,对极几何的准确度是影响边信息的质量以及运动搜索范围大小的关键因素,因此,提出了基于图像配准的匹配点对优化算法以获取更准确的对极几何关系,算法在对特征点对进行坐标优化调整的同时,剔除了部分相关性小的匹配点对,提高了其匹配精度。将该算法所得到的视点间对极几何关系用于分布式多视点视频编码系统中计算空域边信息,并对系统仿真测试。实验结果分析表明:加入了特征点对优化算法后,直接通过在对极线上搜索所获取的边信息质量提升超过了2 d B;若将搜索范围往对极线垂直方向上扩张,在保持相似重建质量的前提下,提出的算法的搜索复杂度降低为原来的10%~20%;与同等条件下的MCII算法相比,性能提升0.2~0.7 d B。

In order to compress the spatial redundancy of distributed multi-view video coding system,a novel method based on the inherent epipolar geometry between different views was proposed to generate the spatial side information. In this process,the accuracy of the epipolar geometry directly impacts motion search range and the quality of side information. An optimization algorithm based on image registration was proposed to obtain a more accurate epipolar geometry,which improved the matching accuracy by optimal adjusting the coordinate of feature points and removing part of feature points with low correlation. The epipolar geometry relationship was used to compute the spatial side information of distributed multi-view video coding system,and its performance was tested. The experiment results showed that by this algorithm,the PSNR of side information increased by more than 2 d B through motion search on the epipolar line. When expanding the vertical motion search range,it could get the similar reconstruction quality with 10% ~ 20% of complexity of the original's. Meanwhile,the rate-distortion performance of the proposed system gained 0. 2 ~ 0. 7 d B compared with the MCII algorithm in the same conditions.