Key Techniques Of Image Mosaics Based On Feature Points

Image mosaics technology has been widely used in the fields of the constructionof virtual scene, analysis of remote sensing data, processing of medical image, andsynthesis of panoramic image, which also has been researched and discussed activelyin the worldwide.At present, it is easily lead to image blurring for the loss of details based on thatthe characteristics of the feature point detection is carried out under the linear scalespace. However, it is rarely researched that make feature detection and description byusing the experience factor diffusion filter to construct nonlinear scale space. In theprocess of matching, RANSAC method is easy to be affected by false match point,which would lead to unsound registration results.This article focuses on research and discussed the current image splicingtechnology advantages and disadvantages, and analyzes the key technology based onpoint feature stitching. To carry out the research work is reflected in the followingfour aspects:Firstly, explore the three kinds of algorithms of SIFT, SURF, KAZE principleand process. To solve the problem of losing building details in the scale space, use thespread of the fitting factor to improve the conduction function, and combine theAdditive Operator Splitting algorithm with conduction function (i.e. the OperatorSplitting, AOS), nonlinear diffusion filtering structure nonlinear scale space for thefeature points location and description. This article verifies the advantages ofimproved conduction function by experiments, compares the three algorithms in theface of different environmental performances, and gives their applicable occasions ofthe three algorithms.Secondly, use the Euclidean distance and the BBF algorithm to search in thedescribed feature points quickly to get the coarse match point. Then, put forward animproved RANSAC adjacent probability method in view of the external point due tothe inefficient problem, which has been proved to be a robust feature matchingstrategy. This strategy effectively remove false match point and ensure to match therobustness and the success rate. Transfer images in the filtered matching point, andestimate the transform parameters of mathematical model and accurately locate theoverlapping area.Thirdly, roughly discusses the advantages and shortcomings of several kinds offusion splicing method and adopts the fading method in view of the stitching lineremoval. The experimental results show that it eliminates the stitching imagesstitching joint problems.Finally, compares the SIFT, SURF, KAZE every performance by using Matlabas a platform for simulation. Using of adjacent probability RANSAC algorithm isproposed in this paper to well guarantee the correctness of the matching results withsuccess.