基于区域先验信息的去遮挡立体匹配算法

De-Occlusion Stereo Matching Algorithm Based on Regional Prior Information

提出了一种基于区域先验信息的去遮挡立体匹配算法。该方法采用边缘检测和区域生长对参考图像进行区域分割,将分割后的同质区域作为区域先验信息引入代价计算中,减弱了代价函数对遮挡区域的敏感度,并使用区域一致性检测修正遮挡和误匹配像素,从而获得精确视差图。所提算法是一种加性算法,在不显著增加计算量的同时,可有效提升算法对遮挡区域的匹配效果。使用Middlebury数据集进行测试,测试结果表明,所提算法能够有效降低遮挡区域的误匹配率。

This paper proposes a de-occlusion stereo matching algorithm using regional prior information.According to this algorithm,the reference image is segmented using edge detection and region growing.The homogeneous regions after segmentation are regarded as the regional prior information and introduced into the cost calculation for weakening the sensitivity of the cost function to occlusion regions.Area consistency detection is used for correcting occlusions and mismatched pixels to obtain an accurate disparity map.The proposed algorithm is an additive algorithm that can effectively improve the matching effect of the original algorithm when applied to occlusion areas without significantly increasing the amount of calculation.The result of testing the proposed algorithm on the Middlebury dataset proves that the algorithm can effectively reduce the mismatching rate when analyzing occlusion regions.