针对光学与红外图像灰度差异大,经典的SIFT算法容易造成误匹配的问题,提出了一种采用互补特征进行匹配的方法。在特征检测阶段,结合两种互补的尺度不变特征检测子DoG和Harris-Laplace来检测团块类和角点类结构,同时为了消除DoG检测子对匹配率的影响,在不影响特征表征性的同时,在更大邻域范围内搜索特征,有效减少了误匹配点数。实验结果表明,文中方法对于多源图像的匹配,能有效增加匹配点数,提高正确率。

In view of the big gray difference in multi-sensor image, and classical SIFT algorithm is prone to cause the error match; a method of adopting complementary features for matching was presented. During feature detection, the DoG detector was combined with Harris-Laplace detector to detect the blob-like and corner-like structure. In order to decrease the error matching caused by overabundance feature of DoG, searching features in bigger areas and the feature distribution was improved. The experimental result shows that the matching points increase effectively and the correct matching ratio is improved.