Dense Matching for DSM Generation From ZY-3 Satellite Imagery

A novel method for dense matching for DSM generation is presented in this paper. First, a global energy function that contains matching cost and smoothness constraint is constructed, and then multiple candidates for disparity of each pixel are obtained by solving the global energy function, where multipath dynamic programming (MDP) is employed. Subsequently, a reliability measure is proposed, which evaluates the reliability of a given candidate and is calculated by Parzen estimation. The candidate with the highest reliability is selected as the final disparity for a given pixel. Moreover, postprocessing steps for removing outliers and the interpolation of gaps are presented to refine the disparity result. Consequently, forward intersection is employed to generate DSM with the disparity result. The complexity of this method is linear to the number of pixels and independent of the disparityrange, which is the reason for generating disparity result in a short time. The experiment conducted on ZY-3 imagery demonstrates the efficiency and accuracy of the proposed method.