Image Registration Algorithm Based On Fuzzy Image Feature

Image registration technology can estimate the geometric mapping relationship and determine the geometric transformation between images under different time,angle,acquisition equipment and environment.Image registration technology is the key to image fusion,super-resolution reconstruction and many other problems.Due to the limitation of the hardware conditions,the device is often collected with blurred and low resolution images.The position of the feature points from the blurred image is often inaccurate and the number of the feature points is large difference between different blur images.Aiming at the problem of the blurred image,a new blurred image registration algorithm based on the blurred image feature points is proposed.The blurred image registration algorithm is a sparse optimization algorithm under the assumption of the rigid and the non-rigid transformation of image.The main innovations of the algorithm include:(1)In the rigid transformation model,the linear sparse optimization model is established by using the blurred image feature points.(2)In the non-rigid transformation model,the linear sparse optimization model is established for the blurred image feature points by using the compact support radial basis function and the elastic network regularized constrained deformation function.(3)Using the KD tree and BBF algorithm to rough registration,the linear sparse optimization model can be affected solved by a fast iterative shrinkage-thresholding algorithm.Alternately matching feature points and estimating image deformation functions.Experiments show that the proposed rigid and non-rigid registration algorithms based on the blurred image feature points have better robustness and better accuracy than the existing RANSAC and M estimation algorithms.In particular,the algorithm that we proposed has better robustness and accuracy in the case of less feature points and fewer feature matching pairs.