一种快速区域立体匹配算法设计及实现

摘要

In order to set up the corresponding relationship between pixels of two homologous images accurately and overcome the shortcomings of large amount computation in method, an improved stereo matching algorithm is proposed. The improved area-based stereo matching algorithm adoptes the two-way stereo matching strategy to improve the accuracy of matching between pixels of two homologous images. The dense parallax error of image is obtained in combination with the mean-shift image segmentation technology under the assumptions of the same area with the same parallax. Meanwhile, in order to conquer the disadvantage of large amount computation of the area-based stereo matching algorithm, the pseudo-epipolar constraint is introduced by this algorithm to limit the location of the true matching point in a very small area. Therefore, the search range of matching point is reduced significantly, and the computation amount of matching algorithm is decreased efficiently. The experiments results show that the improved algorithm can effectively reduce the matching range and achieve the exact dense matching between images.%为了精确建立两幅图像像素间的对应关系,克服区域立体匹配算法计算量大的缺点,提出了一种改进的区域立体匹配算法.改进的区域匹配算法运用双向立体匹配策略来提高两幅图像像素间的匹配精度,并结合mean-shift图像分割技术,在同一区域具有相同视差的假设前提下获得图像的稠密视差;为了克服区域立体匹配计算量大的缺点,该算法引入伪极线约束将真正匹配点的位王限定在一个非常小的范国内,从而大大减少了匹配点的搜索范围,有效地降低了匹配算法的计算量.实验结果表明,改进的双向匹配算法可以有效地缩小匹配搜索范围,并实现图像间的精确稠密匹配.