Research On Image Matching Method For Scene Perception

Image matching is significantly important in image processing which is the foundation of image fusion,object tracking and so on.Current image matching has a mature frame:feature extraction,similarity meature and searching strategy.According to the feature,image matching can be sorted into two classifications: gray-value based image matching and feature based image matching.The two classifications have their own application scenarios.Because of dependent seriously on the gray value,gray-value based image matching may have many wrong matching scenes such as background changing,image rotation,scale,noise and blur.Feature based image matching has many applications because of its invariance.As a study for image matching,this article has done some research as follows:(1)This article has a detailed introduction of frame of image matching:feature extraction,similarity meature and searching strategy.For feature extraction,we introduce color feature,shape feature and edge feature,for similarity,we introduce gray-value similarity,euclidean distance,mahalanobias distance and hausdorff distance.For searching strategy,we introduce three step searching,best bin first and genetic algorithm.(2)Some detail introduction of Scale Invariant Feature Transform(SIFT),Principal Component Analysis SIFT(PCA-SIFT)and Speeded-Up Robust Features(SURF)has been done,the matching algorithms of them are also finished.At last,we compare the difference performance on time,scale,rotation,illumination and blur.SURF is the best algorithm on time,SIFT has better performance with other situations.(3)Based on SIFT and PCA-SIFT algorithm,article propose a new algorithm for image matching based on SURF which doubles the area of feature descriptor.we test its performance with four situation.Before test,the matrix of transform is a known,the matrix can judge whether the pair of matching is true or false.Article use three criterions:time,recall and precision to judge.At last,after comparing the two image matching algorithm,we get the algorithm we propose has better performance.(4)This article designed and realized the software of basis models and algorithms of image processing and computer vision.Software is divided into three parts,we integrate the gray-level based image matching algorithms and feature based image matching algorithms into it,at last,we use this software to check the algorithms we realized.