Research On Adaptive Corner Detection And Image Matching Algorithm For Remote Sensing Image Based On Autocorrelation Matrix

The image matching based on corner features plays an important role in the digital processing of remote sensing images,and the extraction of feature points is the key to image matching.The data of remote sensing image is rich in information.However,due to the large amount of calculation of the traditional corner detection algorithm,the positioning is inaccurate and it is easy to extract the pseudo corners.In this paper,a threshold adaptive corner detection algorithm based on pixel autocorrelation matrix is proposed,which greatly improves the efficiency and accuracy of corner detection.A modified Harris corner detection algorithm based on auto-correlation matrix of pixel is proposed in this paper,and the purpose is to solve the problem of the variability and randomness of the thresholds of corner response function(CRF)and non-maximum suppression in Harris algorithm.First,GF-2 remote sensing image’s multi-spectral data and panchromatic data were fused to a high-resolution color image by orthorectification,automatic matching,and Gram-Schmidt Pan Sharpening method.The corners of the building in the image were important detection targets.Second,the fusion image was filtered by directional filtering and low-pass filtering,and Feature Corner Image(FCI)was constructed by calculating determinant of every pixel’s auto-correlation matrix.Third,the adaptive segmentation threshold of FCI was calculated by using optimal entropy algorithm,and the pre-selected regions were selected.On the basis of the front,a modified non-maximum suppression method was adopted to extract corners effectively.Finally,several common gray based image matching algorithms and feature based image matching algorithms are introduced.Based on the improved corner extraction algorithm,a cross correlation algorithm is selected to match,and the remote sensing matching image is obtained.By comparison experiments with the traditional Harris corner detection algorithm,we could conclude that the improved algorithm could not only calculate the optimal threshold automatically,but also locate the corners more accurately and improve the precision of corner detection.