Image Matching Method Based on Multi-scale Corner Detection

Abstract

A new image matching algorithm based on multi-scale corner point detection is presented to solve the matching difficulty in infrared and visible images. After establishing the Gaussian scale space, firstly, the multi-scale edge of Canny is found in the Gaussian scale space, the curvature scale space is established, then the corner point is extracted, and finally the multi-scale CSS detector is formed. secondly, in order to avoid the gradient flip when building gradient vector of feature points in infrared and visible image, gradient direction angle of feature point neighborhood is limited and the direction is amended by using nearest projection, the main direction of the feature points is obtained from the histogram of the gradient direction, a 64-dimensional feature point descriptors is constructed and normalized; It is shown that in experiments this algorithm can effectively match the infrared and visible images, the matching result has high accuracy in the circumstantialities of rotation, zoom scale and brightness changes.