Research On Image Registration Based On Feature Points

Image registration is a research hotspot in the field of image processing.Designed to fuse two or more different images of the same scene image from different sensors,different angles of view or different time into one image to meet a specific application.Due to the influence of noise and blur,image registration is difficult to obtain accurate conversion functions quickly.The main work of this paper is to study the efficient and robust image registration algorithm.The specific research contents are as follows:1.In order to reduce the feature point pairs of the Scale Invariant Feature Transform（SIFT）algorithm in the image matching,a more accurate image registration transformation function is obtained.This paper proposes a method that can filter out a large number of noise and outliers in the matching process and recover the spatial transformation function between the point sets.The method estimates the transformation between two non-rigid sparse point sets by iterative recovery point correspondence,removes noise and outliers,and obtains more accurate image registration results.In the first step of the iteration,the SIFT feature descriptors establish a rough correspondence.In the second step,the L2-minimizing estimate estimator（L₂E）is used to estimate the transform function.The L₂E estimate filters out noise and outliers in the feature points,and then models the non-rigid transformation functions in the Reproducing Kernel Hilbert Space（RKHS）space.The simulation results show that the L₂E estimation method is robust to feature points containing severe noise,occlusion,outliers,rotation or scale transformation,and can obtain more accurate registration results.2.In order to solve the unsatisfactory effect of the traditional single image restoration algorithm,on the basis of image registration,this paper uses a variety of images to recover a single image by using multiple images.Firstly,the M-estimation is used to register the image,and then the L1 norm is used for image fusion,which improves the robustness of image restoration.At the same time,in order to achieve the fast convergence of the algorithm,this paper proposes an image restoration algorithm based on Conjugate Gradient Descent（CGD）and its improved algorithm by optimizing the search gradient of the descent algorithm.Both methods can be effective.Reduce image fusion time.The simulation results show that the improved algorithm proposed in this paper is faster than the image recovery algorithm based on the steepest descent（SD）method.