Research On The Mosaic Method Of Sequence Remote Sensing Image

Remote sensing images have been applied in many fields,such as agricultural,forestry,geography,marine meteorology,military,environmental protection and so on.In related research experiments,a large area needs to be researched and analyzed.So,in order to expand the area,two or more remote sensing images must be stitched into a wide-view and high-definition image.There are two main purposes for image stitching research: one is the accuracy of image stitching,and the other is the speed of it.There are many problems in the process of stitching,such as lighting transformation,similar interference,image scaling and so on.Aiming at the accuracy and real-time of image stitching,the research in this paper is divided into three stages: preprocessing,feature point matching and image stitching.In these three stages,a method of estimating overlapping areas based on DBSCAN algorithm,the mutual information and Euclidean distance are combined for segmentation matching Feature point method and image mosaic method based on improved RANSAC is proposed individuallyThe main work:1.In the image preprocessing stage,this paper proposes an improved ORB corner detection method based on the characteristics of feature points.The algorithm is improved from the two aspects of the way of extracting the surrounding points and the conditions for determining the feature points.The algorithm can quickly extract the feature points.Then,based on the distribution of image feature points,a method of estimating the overlapping area of images using adjacency graph is proposed.First,The DBSCAN clustering algorithm is used to cluster the feature points.Then use the clusters to construct the adjacency graph of the image,and finally analyze the adjacency graph to obtain the maximum similarity cluster,thereby estimating the overlapping area of the image.2.In the feature point matching stage,this paper proposes a method for fast matching feature points.First use K-nearest neighbor algorithm to quickly match all feature points.Then,large number of mismatched point pairs in the matching resultneeds to be purified.The first step is to use the overlapping area which obtained from the image preprocessing stage to remove these feature points outside the overlapping area.The second step is to find the correct matching point pair,using the Euclidean distance of the matching point pair,the matching point pair will be kept when the ratio of the nearest neighbor to the next nearest neighbor is small.The last step,the binary mutual information of the image is used to judge the feature point pairs with larger ratio.By matching feature points in this way,the amount of redundant calculation is reduced,and the accuracy of matching feature points is improved.3.In the image mosaic fusion stage,this paper proposes an image mosaic method based on the improved RANSAC algorithm.This algorithm is based on the iterative calculation principle of RANSAC algorithm.During the iterative calculation,the feature point samples are dynamically changed to increase the ratio of interior points,thereby reducing the number of iterations.With this algorithm,the transformation matrix of the image can be quickly obtained.Then the affine transformation of the image is realized using the matrix.Next,the stitching traces of the image are eliminated by the Poisson fusion method.Finally,the stitching of serial remote sensing images is implemented by pairing.