视差匹配是中间视合成、立体图像/视频编码及物体识别的关键步骤。提出了一种在大基线下基于对象的预测递归视差匹配算法。首先识别前景和背景区域,然后在对前景对象进行匹配的过程中,采用相邻块预测方法,提高了匹配的速度和精度。实验结果表明,本算法能得出良好的视差图。

Disparity match is the key technology that realizes intermediate view synthesis,multiocular video coding and object recognition.In this paper,a predict recursion disparity match algorithm based on object is put forward.Firstly,foreground area is distinguished from background,and then the disparity of only object area is estimated with information of neighbor block.Experimental results show that this method can obtain disparity map with high quality.