Feature guided multi-window area-based matching method for urban remote sensing stereo pairs

This paper presents a feature guided multi-window area-based matching method for urban remote sensing stereo pairs. The method achieves the goal that producing dense disparity maps for urban remote sensing stereo pairs. The proposed method can be divided into four stages: feature-based matching, edge support region extraction, area-based matching and post-processing. The point feature matching is applied firstly as it provides precise matching results of few points, which can be used to restrict the searching range of area-based matching process. Then, building edges are extracted by an efficient line segment detector. In order to enhance the accuracy of the area-based matching, edge support regions are derived from these edges and multi-windows are applied during the area-based matching to preserve details of edges. Finally the post-processing including interpolation and filtering is applied to heighten the completeness and accuracy of the disparity map.