使用对极几何的多视角分布式视频编码

Multi-view distributed video coding using epipolar geometry

为了降低多视角分布式视频编码中解码器复杂度,解决视角间边信息的获取精度等问题,将一种新的基于对极几何的视差估计算法用于多视角分布式视频编码系统。提出的系统通过对视角间的冗余信息使用对极几何约束,采用新的视差搜索开始点,限制垂直方向的搜索范围以加快搜索速度。实验结果表明,提出的系统相对于运动JPEG(MJPEG)编码的主要运算绝对差值和(SAD)计算次数减少10倍,相对H.263+减少7倍。同时提出系统相对MJPEG编码有6～7dB增益,相对H.263+有3dB增益。

To reduce the complexity of decoder for multi-view distributed video coding and solve the accuracy of accessing side information,a novel epipolar geometry-based disparity estimation algorithm is used for multi-view distributed video coding system.By using constraints of epipolar geometry for redundant information of inter-views,proposed system applies a novel disparity search center and restricts the vertical search space to accelerate searching.Experiments results show that the main operations Sum of Absolute Difference（SAD） of proposed system calculate 10 times less than Motion JPEG（MJPEG）,7 times than H.263＋.Meanwhile,the proposed system gets compression gain 6～7 dB compared with MJPEG,3 dB with H.263＋.