Local stereo matching with side window

Abstract

Although local stereo matching methods have attracted much attention for their promising performance in disparity estimation, the issue of edge-blurring remains a challenge for existing local methods. To address this issue, we propose a novel local stereo matching method based on side window filtering technique, namely LSMSW. In the proposed method, the multiple-window aggregation strategy is adopted, which can adaptively select the window type according to the different texture information. Compared to traditional local methods, our cost aggregation model shows the significant performance in terms of accuracy and computational efficiency, especially for images with complex edges. The experimental results on the Middlebury benchmark confirm our achievements.