异源图像特征点边缘描述与匹配

Edges Description and Matching Algorithm for Different-source Images

针对红外与可见光图像中特征点匹配的难题,提出一种基于特征点邻域边缘的描述与匹配算法。首先采用基于曲率尺度空间的角点检测算法进行特征点提取;再对特征点邻域的边缘进行重组;其次求取特征点所在曲线的法线作为主方向,以避免图像的旋转代价;计算特征点邻域像素点的B-LBP算子的加权分布直方图;然后搜索相同边缘上最近的特征点并计算相应的直方图信息;再对两个直方图进行级联,构造出512维的UB-LBP联合描述子,并将其归一化;最后采用最近邻算法实现特征点匹配。实验结果表明,这两种描述子在红外与可见光图像特征点匹配方面较SIFT算法具有较高的正确匹配率,能够实现两种图像的精确匹配。

A point matching algorithm based on edges of key points region was proposed to resolve the problem of IR and visible images matching. Firstly, the feature points were extracted by the CSS comer detector. Edges of key points region were reconstructed. Secondly, the normal direction of each feature point on the curve was adopted as the main di- rection of the point, making the point descriptor rotation invariant. Thirdly, through calculating the B-LBP weight histo- gram in interesting points＇ neighborhood, the nearest feature point of each extracted one on the same edge was searched and the histograms of edge pixels of the two key points region were constructed. Then a 512-dimentional UB-LBP joint descriptor combining with two histograms was constructed and normalized. Finally, the feature matching was realized via the nearest neighbor algorithm. Experimental results show that the proposed algorithm can match the feature points in the IR and visible images more efficiently than the original SIFT.