Adaptive similarity fusion strategy in stereo matching algorithm

Cost initialization is an important step in both local and global stereo matching algorithms. Different similarity measures may perform differently in a particular location depending on the image structure. Different from the traditional similarity fusion strategies, we present a new fusion strategy which fuses similarity measures adaptively based on their performance on different image regions. The experimental results demonstrate that the proposed adaptive similarity fusion algorithm used in both local and global stereo matching algorithms is capable of providing high-quality disparity maps comparable to other fusion strategies.