摘要： 给出一种改进的立体匹配算法。根据视差场的极限约束，采用图像线分割的匹配方法得到初始视差，由基于初始视差的交叉检测技术和可信度约束提取高可靠度的GCP点。针对视差图中的不可靠点，加入Bilateral Filter的自适应加权中值滤波，在由均值偏移图像分割算法得到的分割区域内，对像素的视差值进行投票，有效遏制局外点，得到更合理的视差图。在Middlebury test set上进行的测试结果表明，该算法具有较好的视差估计精度。

关键词: 立体匹配, 线分割, 交叉检测, 可信度约束, 均值偏移, 自适应加权中值滤波

Abstract: This paper gives an improved stereo matching algorithm. According to the polar constraint of disparity field, it obtains the initial disparity map based on line-segmentation, and uses the cross detection technology and credibility constraints to acquire the high reliability GCP pixel. To eliminate the unreliable disparity, the paper uses adaptive weighted median filter which is based on the Bilateral Filter and votes the disparity in the region segment which is obtained through mean shift algorithm. This strategy curbs the outlier effectively and obtains the more reasonable disparity map. Test result on Middlebury test set shows that the algorithm improves the performance greatly and has better accuracy.

Key words: stereo matching, line-segmentation, cross detection, credibility constraint, mean shift, adaptive weighted median filtering

中图分类号: 

• TP18