Research On Aerial Image Matching Based On SIFT Feature Selection

In order to solve the problem of inefficient matching algorithm due to too many feature points extracted by SIFT(Scale Invariant Feature Transformation)algorithm and low matching accuracy caused by mismatching in the algorithm,a method is proposed to realize aerial image matching,which is to select high-level features to obtain matching sets,and then use different MSAC(M-Estimate Sample Consensus)models to exclude mismatching.Firstly,the features of Gaussian pyramid image are extracted from the second layer,the dot product between the features is calculated,the inverse cosine result of the point product is taken and sorted.Then,the high and low thresholds are used for feature point matching,and the feature whose ratio of the nearest neighbor to the second nearest neighbor is less than the low threshold value is retained.When the characteristics of aerial images change greatly and there are not enough matching pairs using low threshold matching,high threshold matching is used.The matching accuracy of feature matching sets obtained from different thresholds is different.Finally,for the problem of mismatching pairs in feature matching pairs,MSAC algorithm is proposed to select different models based on matching accuracy to filter matching pairs,and feature matching sets with higher matching accuracy from low thresholds are filtered using homography transformation models,and the affine transformation model is used to filter the feature matching sets with high threshold values.Eight groups of medium-sized aerial images(horizontal and vertical pixels between400-1000)were used in the experimental images.The experimental results show that the efficiency of feature extraction is improved to a great extent by reducing the number of feature points extracted from the images.The efficiency of eight images is improved between 67% and76%.In some aerial photographs with high matching accuracy,the use of affine transformation model will result in feature matching pairs concentrating in a certain area,which is not conducive to image post-stitching,fusion and other processing,while the distribution of feature matching pairs is more uniform when using the homography transformation model;in the case of low matching accuracy,the results of affine transformation model are more stable than that of the homography transformation model.To sum up,by selecting high-level features for feature matching and using different models of MSAC algorithm to filter the feature matching set,the matching efficiency and accuracy of the algorithm are improved.