Research On Image Registration Method Based On Improved SIFT

With the development of digital image processing and computer vision technology,image registration has become a very important image processing technology,widely used in remote sensing military,medical image analyzing,image machine vision,pattern recognition and many other important areas.Recently,lots of efficient image registration methods have been proposed by domestic and foreign scholars focusing on image registration for in-depth study.The most representative method among them is“Scale Invariant Feature Transform”,which was proposed by Lowe.SIFT feature keep good invariance property in image rotation,angle transformation,affine transformation and scale zooming.Feature matching as the most important part of the image registration,has always been emphasis that the scholars are studying on.The main research work in this paper is aimed at realizing feature points accurate searching and effective matching based on improved SIFT algorithm,the main contents of this thesis are as follows:(1)The background of image registration and the research status at home and abroad are introduced in detail,and the principle of image registration and the basis of mathematical theory are given.Several common geometric transformation models are introduced,which is providing theoretical basis for subsequent image registration.(2)SIFT algorithm principle is discussed in detail,for the original SIFT method has certain false matching points,misses a large number of correct matching points and the correct matching rate is low in image registration processing,an improved SIFT algorithm based on scale,orientation and distance constraint is proposed in this paper.We remove false matches through adding constraint factors.Experimental results demonstrate that the proposed method can obtain more correct matches,and the correct matching rate is improved.(3)Aiming at the problems of the number of matches is less and the correct matching rate is low during remote sensing image registration processing,an improved sift method based on local-affine constraint was proposed in this paper for remote sensing image registration.We presented a new gradient operator,and feature descriptors are constructed through circular neighbourhood instead of square neighbourhood,then we achieved feature matching through fast sample consensus algorithm and affine transformation local searching algorithm.The experimental results show that our method has good performance on remote sensing image registration,thecorrect matching rate and accuracy of image registration are both improved.