Towards Stereo Matching Algorithm Based on Multi-matching Primitive Fusion

Classical adaptive support weight (ASW) algorithm has poor robustness and high computational complexity for stereo matching in the case of relatively low texture and complex texture regions. To solve this issue, a novel stereo matching algorithm based on the multi-matching primitive is proposed by combining color matching primitive with gradient matching primitive and integrating the correlation. This algorithm consists of three stages: initial matching cost stage, aggregation stage of cost function and parallax post-processing stage. In the first stage, we design a cost function incorporating color primitives and gradient primitives. In the second stage, we develop an adaptive matching window based on the relationship between RGB color and the space distance. In the last stage, we perform parallax post-processing by Left-Right Consistency check and adaptive weight median filtering based on Sub-Pixel. Experimental results showed that the proposed algorithm has good performance in the case of low texture and complex texture regions compared with ASW.