Implementation Of Image Corner Extraction And Research Of Image Matching

In recent years,computer vision technology is developing rapidly,image feature extraction and matching as the key techniques of the digital image processing field,is getting widely used,especially in the field of military guidance,remote sensing aerospace,target detection and tracking,three-dimensional reconstruction,medical image analysis,image retrieval,etc.It is playing an increasingly important role.Image matching is to match the image which is under different imaging conditions with original image of the same scene.Among the many image feature,corners are able to contain important information in images,maintain image edge stability,and also have the advantages of rotation,scaling,displacement and matching simplicity.Therefore,corners become one of the important features of image matching.This paper focuses on image matching technology,first on the basis of studying Harris corner extraction algorithm,proposes an improved algorithm for comparing the pixel gray value in the neighborhood before calculating the response value,to remove some non-corner,and the algorithm is realized by FPGA,using the DE2-70 development platform,input the image pixel value stored in RAM memory into the FPGA processing module by the way of address strobe,and output the coordinates of image corner pixel.The whole system includes two modules of corner extraction and corner registration.The corner extraction module realizes Gaussian filtering,Harris response value calculation and non-maximal suppression,the corner registration module realizes the function of corners' coordinates calculation and registration.Simulate the algorithm processing unit in the Modelsim,save the corners' coordinates to the txt file,and call these corners through MATLAB to finish registration,show and analyze the results.And then,this paper studies an image matching algorithm based on Harris-SIFT feature extraction,makes up the shortcomings of Harris and SIFT feature extraction,and optimizes the original 128-dimensional feature description by using a new 64-dimensional feature description.The feature matching part adopts the bidirectional Euclidean distance to measure the similarity,search strategy chooses BBF search algorithm,search efficiency is higher.The simulation of the improved algorithm based on Harris-SIFT feature extraction and matching is carried under different conditions,the results show that the algorithm is superior to the classic SIFT matching algorithm both in matching performance and matching efficiency,and the algorithm can adapt to the rotation,noise interference and other changes well,and has certain applicability.