High-Resolution Satellite Stereo Matching by Object-Based Semiglobal Matching and Iterative Guided Edge-Preserving Filter

Abstract

This letter presents an object-based stereo matching and content-based guided disparity map refinement for high-resolution satellite stereo images. The proposed method uses the superpixels for object-based cost filtering and then object-based semiglobal matching. In the proposed object-based stereo matching strategy, two improvements including homogeneity weight for cost filtering, and weighted cost aggregation by image objects are considered. After cost aggregation, the generated disparity map is computed using the winner takes all and then refined with a new iterative guided edge-preserving filter. The proposed method with Census cost function has been implemented on high-resolution satellite stereo images and then is compared with the LiDAR ground truth. Moreover, the proposed method is compared with five state-of-the-art stereo matching methods. The experimental results on along-track satellite stereo images from Pleiades and IKONOS images and cross-track multidate images from WV-III demonstrate that the proposed method significantly improves the result of stereo matching.