Abstract:

A point matching algorithm based on the curvature scale space (CSS) corner extraction is proposed to resolve the problem of point matching in the infrared (IR) and visible images. Firstly, the candidate feature points are extracted with a CSS corner detector. Secondly, the low contrast points are rejected by the 3D quadratic function. And thirdly, the normal direction of each feature point on the curve is adopted as the main direction of the point, thus the rotation of images is avoided. Fourthly, the gradient histograms of each keypoint region are constructed, and the weighted and normalized to unit norm statistic characteristics are calculated to build a 64D descriptor. Finally, the feature matching is realized via the nearest neighbor algorithm. Experimental results show that the proposed algorithm can match the feature points in the IR and visible images efficiently and correctly.