【摘要】 文中提出了一种新颖的基于窗口的立体匹配方法,该方法首先在最大窗口内估计视差,并假设该窗口内视差一致,在此基础上得到两个最大的匹配窗口,然后在这两个最大窗口内进行二次匹配,得到基于这两个窗口的逐像素视差,估计视差和像素视差之和就是结果视差。对相对较平滑或平滑均匀的区域,在匹配过程中,会出现匹配最小多值问题,面临如何确定最佳匹配。本文算法采用平滑性测度指标函数来屏蔽平滑或平滑均匀区域,并在匹配完成后,按照最近邻视差均值来估计平滑区域的视差。最后通过立体图像对算法进行了测试。实验结果表明,该方法是可行和有效的。更多还原

【Abstract】 A novel window-based stereo matching approach is proposed in this paper.The approach estimates the coarse disparity in maximum window in left and right image according most-min absolute difference method firstly.As result, two maximum windows are established,after then,the second match is done in the two max windows and obtain fine disparity,the result disparity is the sum of coarse disparity and fine disparity.For the smooth or wipe regions,the problem of cost function’s multiple most minimums is appeared in the process of stereo matching.In order to overcome it,threshold smooth cost function is presented to avoid the smooth or wipe balance regions.After completing the matc- hing of no-wipe or non-smooth regions,the unmatched wipe and smooth regions are estimated by the average of most adjacent disparity.At last,several stereo images are used to test the algorithm.The results indicate that the approach is valid and possible.更多还原