A Complementary Aggregation Approach for Local Stereo Matching Using Color and Correlation Cues

Existed local stereo methods usually choose the support region for aggregation using correlation or color information independently. The correlation cue works well with high textures but has a poor performance near depth discontinuities while the color cue plays the complementary role. In this paper we first propose a new soft segmentation approach for correlation-based aggregation. Then we make a combination of the two cues and adopt the advantages of them to overcome the limitation of each other. Our approach performs a two stage aggregation based on correlation and color respectively. Each stage is operated by a bilateral filter on the cost volume. The combination is simple and effective, which enables our approach to achieve a better performance in both highly textured areas and depth discontinuities than existed methods. The experimental results conform to our expectation and do make improvements to state-of-the-art methods.