Minimum spanning tree dynamic programming stereo-matching method based on superpixels

The minimum spanning tree (MST) stereo-matching method is an information-infiltration process. The difference in edge attributes of an MST can cause the edge-expansion phenomenon, which affects the matching accuracy. To accurately recover image-depth information, a dynamic-programming stereo-matching method based on the MST was proposed. First, the colour Birchfield Tomasi cost-calculation method based on image adaptive colour information was proposed to obtain stable initial cost values. Second, the image was segmented into superpixel regions using the simple linear iterative clustering algorithm. The pixel- and region-level MSTs were then constructed. Next, combined with the idea of dynamic programming, the MST cost-aggregation process was re-deduced. On this basis, the aggregate cost values of the two MSTs were obtained. Finally, the aggregate cost values were combined adaptively to acquire the high-precision smooth disparity map. The Middlebury 2014 dataset was used for the experiments. The experimental results indicate that the proposed method can effectively improve the accuracy of stereo-matching.

Automated summary

This study proposes a dynamic programming stereo-matching method based on the minimum spanning tree (MST). By obtaining stable initial cost values, constructing pixel- and region-level MSTs, and re-deducing the cost-aggregation process using dynamic programming, a high-precision smooth disparity map is obtained. Experimental results demonstrate the effectiveness of the proposed method in improving the accuracy of stereo-matching.