BFAST_CSP_KAZE computing model for effective image registration

Abstract

Abstract The incompatible problem between velocity and accuracy has been restricting the application and the development of image registration, in order to resolve this problem, we propose the BFAST_CSP_KAZE computing model. This model consists of four stages. First, the registration images are preprocessed by the average and the perceptual Hashing algorithms. Second, the additive operator splitting algorithm is used to construct the nonlinear multi-scale space and utilize the FAST algorithm to extract the image features. Third, a new method to generate the image features descriptor sequences is presented based on the optimized KAZE algorithm, meanwhile, reduce the dimension of the image features descriptor sequences by the compressed sensing principle. Finally, adopt the fusion matching method based on the FLANN and the KNN algorithms to match, and the RANSAC algorithm further match. This paper utilizes two groups of the experiments to verify this model, the experiment results show that this model has obvious advantage in terms of velocity and accuracy compared with the state-of-the-art image registration methods, and also achieves the compatible between velocity and accuracy in the case of the higher matching score. This model provides an effective solution for the application of image registration, and also has great significance for the development of image registration.