融合Spearman相关性系数与多尺度框架的立体匹配算法

针对现有立体匹配算法对噪声敏感、匹配率低的问题,提出了一种基于Spearman相关性系数与多尺度框架融合的立体匹配算法。在代价计算阶段,创新性地在固定窗口内通过简化Spearman相关性系数得到两种代价计算模型。在代价聚合阶段,利用多尺度框架在图像金字塔上进行代价聚合,从而使得匹配算法在低纹理区域得到较高的匹配率。实验结果表明,提出的立体匹配算法有效降低了误匹配率:对Middletury2.0测试集中31对标准图像对的平均误匹配率仅为7.98%,Middletury3.0中的15对标准图像对的平均误匹配率为13.45%。实验结果表明,提出的融合Spearman相关性系数与多尺度框架的立体匹配能有效降低图像的误匹配率,并对噪声等具有较好的稳健性。

An algorithm that fuses Spearman correlation coefficient and multi-scale framework for stereo matching

Aiming at the noise-sensibility and low matching rate of existing local matching algorithms, a stereo matching algorithm based on Spearman correlation coefficient and multi-scale framework was proposed. Two cost calculation models were proposed by using a fixed window and by simplifying the Spearman correlation coefficient. Then, a multi-scale framework was fused to perform cost aggregation on the image pyramid in order that the matching algorithm could obtain a higher matching rate in the low-texture region. The experimental results show that the proposed stereo matching algorithm effectively reduces the false matching rate: the average mismatching rate of 31 standard image pairs in the Middlebury 2.0 test set is only 7.89%, and the average of 15 standard image pairs in Middlebury 3.0 is 13.45%. Therefore, the method can effectively reduce the mismatching rate of images and has better robustness against noise.