Suitable Matching Area Selecting Method Based On SIFT Feature

Choosing the suitable matching area is one of the important tasks of the scene matching system,and it is the premise and basis of the track planning.The feature distribution of the matching area directly affects the matching probability and matching accuracy.In order to make the guidance system accurately and reliably locate and reduce the navigation deviation,it is required to select the scene features to satisfy the rich,stable and unique area as the matching area in the track planning.Therefore,it is necessary to analyze the matching probability of each candidate area by using appropriate image features.In turn,the suitable matching characteristics of the region are obtained.The paper takes the SIFT feature points of the scene region and its feature descriptors as the initial feature set of the region,and supports vector regression as the matching probability prediction model.By threshold segmentation of the prediction probability of the scene region,the suitable matching characteristics of the region are finally obtained.The main work of the thesis is as follows:In terms of image feature selection,we use the SIFT features with illumination,rotation and scale invariance as the basic feature quantities of the image of the region.In order to improve the image region information representation ability of the SIFT feature point set,a discrimination is proposed for the SIFT feature stability and unique criteria.For the set of SIFT feature points filtered by stability and uniqueness,the several suitable indexes are proposed based on their richness and uniformity of spatial distribution,and verified by experiments.On the one hand,considering that the real-time image is taken by the aircraft,it will be affected by the flying height,the speckle noise and the image size scaling.As a result,some feature points in the reference image area cannot find the matching point pairs in the corresponding positions in the real-time image,and the SIFT features are needed.The stability of the point is filtered;on the other hand,there may beSIFT feature points with high similarity in the reference image area.In this case,the probability of mismatching is very large,so it is necessary to filter according to the uniqueness of the SIFT feature points.In order to measure the performance of suitable matching of the reference image scene by the characteristics of the matching characteristics,the paper uses the support vector regression model in machine learning to establish the mapping from the characteristics of the matching region to the matching probability,so as to predict the scene.The matching probability of the region,and then the threshold matching to obtain the suitable matching characteristics of the candidate region.The experimental results show that compared with the traditional statistical feature-based adaptive region selection method,the suitable matching region selection method proposed by the paper has higher accuracy.