Research On Matching Method Of Optical And SAR Images

With the rapid development of remote sensing technology,the types of remote sensing images have gradually changed from single to multiple.Due to the different imaging properties,multi-source remote sensing data can obtain differentiated image information.The joint application of multi-source remote sensing data has become an important research point in the field of remote sensing,and the accurate registration of multi-source images is an important prerequisite.The problem of optical and SAR images matching is even more difficult.This paper analyzes the different imaging characteristics of two types of images for the purpose of accurate registration of optical and SAR images.Starting from two types of matching methods,feature matching and template matching,a rough-to-fine matching method is proposed which can obtain reliable corresponding points between the optical and SAR images in the typical area.It achieves accurate registration of the two types of images.The main research contents and innovations of this article are as follows:1.A feature matching method of visible light and SAR image based on ratio of exponentially weighted averages is proposed.This method is aimed at the problem that traditional gray-scale difference gradient is sensitive to the multiplicative speckle noise on SAR images.The ratio of exponentially weighted averages is used for gradient calculation.Feature corners are extracted as points to be matched in multi-scale Harris space,and feature points are filtered by the distribution of scales and grids to avoid threshold selection problems.Considering the problem of gradient direction reversal which is common in multi-source images,an angle limit was added to the assignment of the main orientation.In order to improve the uniqueness of the descriptor,a GLOH operator is used to calculate the descriptor,and a136-dimensional feature descriptor is generated for each feature point.In the feature matching process,a combination of the nearest sub-Euclidean distance ratio and dual matching is used to eliminate outliers at the maximum.Finally,random sampling consistency is used to obtain the final matching result.Several sets of experiments have proved that this method has better performance than the traditional SIFT method and SAR-SIFT method,and can provide reliable initial matching point results.2.A optical and SAR image template matching method based on directional phase features is proposed.Aiming at the lack of feature extraction accuracy of the feature matching method,coarse matching results are used as input,and significant rotation and scale differences between images are eliminated through geometric correction.And then build a feature template to search by pixel by pixel to determine the best matching point.This method is based on the principle of phase congruency,and uses orientation-selective log Gabor filters to extract phase features in different directions.A dense feature template is constructed through the combination of neighborhood directional phase features.Finally,the cross-correlation function is used to find the best match in the search window.Fit the peaks of the correlation function to improve the matching accuracy to sub-pixel.Several sets of experiments have proved that this method can obtain stable and reliable corresponding points,significantly improve the single-point positioning accuracy based on rough matching,and finally achieve accurate registration of optical and SAR images.