基于色彩分割和自适应窗口的快速立体匹配

针对现有立体匹配算法难以兼顾匹配精度和速度的不足,提出了一种基于联合匹配代价的局部方法.首先,根据视差在同一色彩分割区域内平滑变化的假设,提出了一种利用参考图像和目标图像的色彩分割信息获得的基于任意形状和大小支持区域的匹配代价;然后在RGB色彩空间中,通过由窗口内的平均匹配误差、误差方差及较大窗口的偏向误差构成的窗口选择评价函数,获得基于自适应矩形窗口的匹配代价;最后,将这两种匹配代价进行有机的结合构成联合匹配代价,并通过局部优化方法获得稠密视差图.采用Middlebury dataset进行的实验结果表明,本文算法不仅可以提高视差不连续区域和低纹理区域的匹配精度,而且获得的视差与当前主流算法具有可比性.此外,所提算法的处理时间较之当前优秀的局部方法提高了约19～35倍.

Fast stereo matching based on color segmentation and adaptive window

Exiting strveo matching algorithms have difficulties to simultaneously meet accuracy and speed requirements.A novel local method based on joint matching cost is presented.Firstly,according to the assumption that disparity varies smoothly within the same color segment,a matching cost based on the support region with arbitrary shape and size is proposed,which is obtained from the color segmention information of reference image and target image is proposed.Secondly,according to the window selection evaluation ...