| The study of image similarity plays an important role in the fields of target tracking,image matching,and image splicing.The similarity of images is generally based on the global features of the image to determine the similarity of the two images,but there will be inaccuracies.In order to solve this problem,we use the local feature of the image,namely the corner feature to judge the similarity of the image.This paper is based on the image corner feature to determine the similarity of the image.First,Harris algorithm is used to extract the corner features of the image.Based on the original algorithm,the corner detection efficiency is improved by reducing the number of corner response function operations.The corner points of the extracted corner points are then matched by the NCC algorithm.Since the original NCC matching algorithm is used to perform bi-directional search for corner points,in order to improve the accuracy of matching,this paper selects the threshold to improve the accuracy of corner matching.Finally,using the Ransac algorithm to fine-tune the corner points.For the similarity representation of images,this paper uses two methods to calculate the similarity of images: First,calculate the similarity of the image by the ratio of the logarithm of fine matching corners and the logarithm of coarse matching corners;On the basis of matching,the similarity of Euclidean distance is calculated;finally,the total similarity is calculated by weighting.This method not only takes into account the matching pairs of images,but also considers the similarity of corner features between images.In this paper,the similarity of images is divided into different levels.Based on the method given in this paper,the experiments of basic images and experiments of synthetic images are performed.The experimental results show that the results calculated by this algorithm are consistent with the subjective judgments of human eyes and can The similarity of the image is judged scientifically. |
PDF Full Text Request