Multi-Sensor Image Registration Based On Region Of Interest

Multi-sensor image matching technology is the foundation of multi-source remote sensing data fusion,scene matching navigation and some other technologies.Used in this technology,the optical image is more suitable for the human visual perception and easier to interpret,while the SAR can be applied all-day and all-weather with the ability of penetrating the surface and the cloud.The combination of the two can effectively exert their respective advantages and complement each other.However,the different imaging modes of two types of images result in great differences in grayscale and geometric characteristics,which brings challenges to image registration.At present,many scholars have carried out research in this area and achieved some results,but limited by the huge differences in grayscale and geometrical characteristics between multi-sensor images,their performance is not good enough in accuracy and reliability.Therefore,it is very necessary to study the multi-sensor image matching method with high precision and stability.This paper aims to match the optical and SAR images with the landmark features.In order to solve the problems of the existing algorithms,this paper adopts the “step-by-step” matching strategy,which is based on the ROI features extraction algorithm in SAR image,and achieves the fine registration between optical images and SAR images progressively.The specific research content of this paper is as follows: 1.Line feature detection method based on GGS-WHT hybrid model.Line feature is an important visual perception information in the image,which is rich in information and stable compared to other features.There are a large number of robust line detection algorithms in the field of optical image processing,but the intense speckle noise in SAR images makes these algorithms cannot apply to SAR images directly.Based on the Hough transform,the GGS bi-windows is used to suppress the speckle noise and calculate the image gradient,thereby extracting image edge features for parameter accumulation.At the same time,the weight value of the Hough transform cumulative process is changed,and the influence of the gradient on the cumulative result is taken into account.2.Multi-sensor image rough matching method based on the region of interest.Due to differences in geometric features of multi-sensor images,the matching effect is susceptible to interference.Therefore,extracting the region of interest in the image in advance and then focusing the image processing on the region is an effective way to reduce the complexity of the matching algorithm.In this paper,according to the line features and target prior knowledge,the approximate location of the region of interest can be determined.The complete extraction of the region of interest is achieved by the method of region-like growth,which can effectively avoid the adhesion of other features.Then,according to the edge information of the region of interest,the corresponding position of the optical image is searched to realize the rough matching between the optical and SAR images.3.Multi-sensor image fine registration method based on biological vision principle.Due to the difference of geometrical characteristics between optical and SAR images,the image edge features are poorly analogous,so the accuracy of the coarse matching results cannot meet the engineering requirements.For this reason,this paper overcomes the inconsistency of multi-sensor image feature extraction by extracting feature in one image and searching for the corresponding point in another image.Finally,the Delaunay triangulation is used to constrain the geometric relationship between points,thereby removing abnormal points and ensuring the reliability of the algorithm.In the process of searching for corresponding points,Gabor filter is used as a descriptor according to the principle of biological vision,which can solve the problem that the feature matching measure is difficult to determine.