Research And Application Of Image Registration Technology Based On Improved SIFT Algorithm

With the rapid development of computer technology,image registration has become a research hotspot in the field of image processing.It has been widely used in medical,remote sensing,military and computer vision fields.With the further application,higher requirements are put forward for image registration: adaptability,timeliness and accuracy.At present,there are a lot of image registration algorithms,and many research results have been achieved,but there are still some problems need to be solved and improved.So it is necessary to carry out in-depth study of image registration algorithm.This paper focuses on the SIFT algorithm based on local features,the purpose is to improve the real-time and accuracy of the registration algorithm.The main research contents include the following aspects:Firstly,this paper systematically studies the basic theory and key technology of image registration.Starting from the research status of image registration,analyzes the basic framework of image registration,studies four kinds of transformation models and three kinds of resampling techniques,and then evaluates and analyzes the performance of the algorithm from four aspects: complexity,robustness,registration accuracy and registration time.Secondly,this paper studies the basic principle and implementation process of SIFT algorithm in detail.Aiming at the problem of high complexity and poor real-time performance in the feature extraction,an improved SIFT algorithm based on ORB is proposed to reduce the time complexity of the original algorithm and improve the real-time performance.In the registration phase,an improved Kd-tree nearest neighbor query algorithm is used to reduce the time consumed in the "backtracking" operation,and combined with the RANSAC algorithm to reduce the matching pairs and improve the accuracy of the algorithm.The improved algorithm and the original algorithm are compared and analyzed in three aspects: angle change,scale change and rotation change.From the experimental data,the improved algorithm has been improved in terms of registration speed and registration precision,which proves the feasibility of the improved algorithm.Finally,this paper applies the improved algorithm to the splicing of scene images.The method of contrast enhancement and histogram equalization is used to preprocess the image,and then applies improved algorithm to the registration of scene images.In the image fusion stage,analyzes with three methods of eliminating the splicing gap,and found out that the weighted smoothing algorithm can realize the smooth transition of the overlapping region,and then get the actual splicing effect graph.