Research On Feature Extraction And Feature Matching Algorithms In Image Registration

In recent years,with the rapid development of digital image processing technology,image registration technology has become a very important step in the field of computer vision and image processing.The processing results of image correction,image mosaic,image fusion and change detection are directly affected by the quality of registration effect.At present,image registration technology is still not mature,and how to achieve fast and high precision image registration is still a major challenge.In order to complete the research work of image registration technology better,the various steps involved in image registration are deeply investigated.Feature-based image registration is the focus of research.The basic theory and research actuality of image registration technology are described,and the various links involved in image registration are systematically studied in this paper.Meanwhile,how to improve the speed and accuracy of image registration is the focus of research,and finally the image registration is realized.The main research contents and contributions of this paper are as follows:(1)An image registration method based on improved SIFT is proposedIn remote sensing images,enough of the correct corresponding points are hard to find due to the significant difference in the grayscale mapping.At the same time,the number of pixels in the remote sensing image is generally large.Therefore,it is difficult to satisfy the requirement of real time due to the large amount of computation using the classical method of scale-invariant feature transform(SIFT).In this regard,the feature extraction part of SIFT is improved in this paper.To overcome the differences of the intensity map,a new method of gradient calculation is adopted and a new description method is proposed.Besides,in order to reduce the computational cost of image registration and improve the registration speed,feature detection is carried out from the second group of gaussian difference pyramid.Then the method of Euclidean ratio matching is used for feature matching.At last,the wrong matching results are eliminated by fast sample consensus(FSC)algorithm.The experimental results show that,compared with several existing classical methods,this method has better performance in terms of registration time and accuracy,while ensuring the correct realization of image registration.(2)A matching method based on slope and distance ratio constraints is proposedAiming at the problems of time-consuming and error-matching of feature points in the process of image registration based on SIFT,a feature point matching method based on the constraint of slope and distance ratio is proposed.Firstly,SIFT algorithm is used for coarse matching of feature points.Then the coarse matching results are screened according to the matching slope.And finally,the screening results are purified by the distance ratio constraint.Experimental results show that the method has good robustness and accuracy,and effectively reduces the time of matching process.