Similarity Measurement Of Image Feature And Its Application

Image matching is an important image processing technology.It matches the same scene images obtained by different sensors or the same sensor in different time,different perspective and different environment under certain similarity criteria,and determines the geometric transformation relationship between them.Similarity measurement is the criterion of measuring the similarity between image features,it plays a key role in image matching and directly affects the validity and correctness of matching results.Firstly,this thesis researches the three elements of image matching: image features,similarity measurement and search strategy,then discusses the traditional image matching algorithms,and designs a performance evaluation system.On this basis,a fast matching algorithm is proposed by combining L1 and L2 norms as similarity measurement for topographic images.The experiment proves that the proposed algorithm effectively improves the matching efficiency of topographic images.Finally,the software platform of image matching algorithm and performance evaluation system is built.The main contents and contributions of this thesis are summarized as follows:(1)This thesis deeply researches and analyzes the key technology theory of image matching: image features detection,similarity measurement,search strategy and geometric transformation model solving.Among them,similarity measurement includes distance similarity measurement and similarity measurement based on image correlation,and the performance of feature matching based on different similarity measurement is tested and analyzed.(2)This thesis deeply analyzes and discusses the traditional image matching algorithms: template matching based on global features,point matching based on local features and matching algorithms based on other theories,then designs a performance evaluation system,which applies similarity measure to the evaluation of matching performance.(3)A fast matching algorithm based on L1+L2 norm is proposed for topographic images.This algorithm uses the combination of FAST and SIFT features to replace the SIFT features,and unions L1 and L2 norms as similarity measurement.The experiment proves that the algorithm can effectively improve the computing efficiency on the premise of guaranteeing the matching accuracy.(4)The software platform of image matching algorithm is built.The platform is divided into two algorithm modules: image matching and performance evaluation,and image matching is divided into four sub-modules: feature detection,feature combination,similarity measure and image matching.Through this platform,we can test different image matching algorithms.The thesis have completed the coding,transplantation,performance evaluation and platform testing of different similarity measurement algorithms.