Image Matching Based On Improved SURF Algorithm

The definition of image matching is to find out the mapping relation among multiple picture under the same scene.The mapping relation should depend on the structure,content,feature,gray level and vein,etc.After all,different images could match under the same space,and this is how could we achieve the image matching.The background of this paper is to locate the position of drone when there is a signal interference.To do that,we need to apply aerial image from the drone and satellite map.The main research of this paper is to accomplish the accurate matching between the aerial image and satellite map.Choosing image matching algorithm with better performance and designing better program according to the feature of aerial image will be the key points for the image matching.Improving feature extraction process of SURF algorithm,improving the accuracy of image matching,providing an optimized solution for satellite map and providing an optimized solution for the aerial image is the significant four parts for this research paper.For the feature extraction process of algorithm,this paper states to apply HSV color space to extract color information of image,and put this color information into descriptor.This change will show the difference and peculiarity among descriptor vectors of feature points.This paper also applies the K-Nearest Neighbor which does the general extraction to other nearby feature points and enhances the accuracy of image matching by removing the worst feature point pair.Then this paper applies the Random Sample Consensus to estimate sample model.This enhancement will decrease the matching error rate.The result clarifies those improvements can obviously enhance the accuracy rate of image matching.In order to solve the problem of giant satellite image,this paper points out to separate one image into multiple smaller pictures.Firstly,we separate the satellite map into several parts,then we use those small parts to match with the separations of aerial image until we get the best performance of the image matching.If the resolution ratio is different between the map and aerial image,then the scale zooming will be the solution of this situation.The more optimized scale will get a batter performance.All of the experiments from this paper can improve the precision of image matching and narrow down the time for matching.