Abstract:

To solve the problem of mismatching in the textureless and edge regions of the non-local algorithm based on minimum spanning tree, an improved cost measurement and color and edge fusion non-local stereo matching algorithm is proposed. Firstly, the cost function based on color-gradient is reconstructed to improve the matching effectiveness in textureless regions. Secondly, an adaptive edge weight function based on image color and edge information is proposed, and the weight is used to construct the tree structure for cost aggregation. The final disparity map is obtained after disparity computation and non-local disparity refinement. The experimental results on the Middlebury dataset show that the proposed algorithm achieves good matching results in both the textureless and edge regions, and improves the disparity effectively.

Key words: stereo matching, minimum spanning tree, textureless region, edge information

摘要：

针对基于最小生成树的非局部算法在无纹理以及边缘区域出现误匹配的问题，提出了一种改进代价计算和颜色与边缘融合的非局部立体匹配算法。首先重新构造了基于颜色-梯度的代价计算函数，以提高无纹理区域像素对代价聚合的贡献率；其次利用颜色与边缘信息进行融合来构造自适应边权函数，并利用该权重构建树结构进行代价聚合；最后通过视差计算和非局部视差优化得到最终的视差图。在Middlebury数据集上进行了测试，实验结果表明，提出的算法在无纹理及边缘区域都取得了良好的匹配效果，有效地改善了视差。

关键词: 立体匹配, 最小生成树, 无纹理区域, 边缘信息