Fusing color and texture features for stereo matching
Yaolin Hou, Jian Yao, Bing Zhou, Yaping Liu · 2013
This paper presents a novel method for stereo matching by fusing color and texture features. Instead of using color features only, which is very similar to most of stereo matching algorithms proposed in recent years, we propose to integrate color features and texture ones in the matching cost function. In this way, more reliable matches can be found in the stereo matching. Firstly we introduce a new texture descriptor, namely, the multi-layer Local Ternary Patterns (LTP), in the matching cost function. Different from the original LTP, our multi-layer descriptor can represent the pixel as specific as possible. Secondly a recently proposed filtering method called the adaptive manifolds filtering is adopted as the cost aggregation strategy. This filtering is a high-dimensional real-time filter operator, which has been verified to be very effective in many image processing fields while is firstly utilized in stereo matching. Finally experiments conducted in the Middlebury benchmark sufficiently verify the effectiveness of our proposed stereo matching algorithm.