Research On A Local Features Matching Algorithm With Robustness In Illumination

Image matching algorithm based on local feature points is an important part of image processing direction such as SLAM system,video stitching and image stitching,so it has been widely studied by scholars at home and abroad.Image matching algorithms based on local feature points need to satisfy affine invariance,scale invariance,compression ratio invariance,illumination robustness,etc.due to differences in image capture conditions,image storage formats,and illumination intensity during shooting.The same image matching algorithm cannot be applied to all scenarios,so it is necessary to improve the image matching algorithm in a specific scenario to increase the robustness of the algorithm.The influence of factors such as illumination difference and camera imaging difference on the image will cause the difference in image brightness to cause the stability of the image matching algorithm based on local feature points to decrease.This thesis mainly studies how to increase the illumination robustness of the matching algorithm.The image matching algorithm based on local feature points is divided into three steps:image graying,feature point extraction,descriptor establishment,and analysis and summary for each step.Aiming at the shortcomings of existing matching algorithms,a local feature point matching algorithm with good illumination robustness is proposed.Firstly,the image is grayed out by RTCP algorithm,and the image information after graying is retained as much as possible.Then,the influence of different illumination on the image is simulated by means of Sigmoid response function,and the feature points with illumination robustness are extracted in the multi-layer image;The local intensity order pattern is used to extract feature point local information to establish descriptors;the Euclidean distances of different descriptors are calculated,and the similarity between descriptors is measured in turn.Experiments show that with the increase of image brightness difference,the accuracy of the scale invariant algorithm,the acceleration robust algorithm,the KAZE algorithm and the ORB algorithm decrease rapidly,and the proposed algorithm is slow and the correct rate is higher.80%;the proposed feature point detection is slower and the descriptor dimension is higher,the average time is 23.47 seconds,the matching speed is not as good as the other four algorithms,but the matching quality is far more than them.In the system with low real-time requirement,the proposed algorithm can weaken the influence of illumination change on image matching.