An Infrared and Visible Image Registration ...

In view of the slow speed of SIFT (scale invariant feature transform), a method of image registration based on SURF (speeded up robust features) was proposed. Firstly, interest points were detected by using SURF and the dimension of the descriptor was 64. Then we selected Euclidean distance as the matching method. Due to the error matching, an improved RANSAC (Random Sample Consensus) was proposed to get rid of them. For their difference, infrared and visible image was preprocessed by edge extraction to improve the similarity between them. Experiment results show the effectiveness of this method.