GEOMETRIC STRUCTUREBASEDIMAGE CLUSTERINGAND IMAGE MATCHING

images ofn objects, thetaskofimage clustering isto We proposetwo geometric structure basedseekan unsupervised algorithm thatcan groupthe approaches GGCI (global geometric clustering for images into ndisjoint subsets suchthat eachsubset only image) andGSIM(geometric structure basedimage contains images ofasingle object. Asanexample, we matching) forimageclustering andimagematching, consider theproblem ofclustering faceimages taken respectively. Forfaceimages orobject images taken with withvarying pose,expression, eyes(wearing sunglasses varying factors, theGGCIapproach learns theglobalor not) or object images underdifferent viewing geometric structure ofimages space andclusters imagesconditions, thegoal istogrouptheimages ofthesame based ongeodesic distance instead ofEuclidean distanceindividual together. Inmanycasesofinterest, these andtheextended nearest neighbor approach. TheGSIM images arefound tolie onanembedded sub-manifold of approach usestheminimal Euclidean distance betweenthehigh dimensional space. Inthis form, we try todivide parts ofimageandthepattern anditsvariations as thesedifferent manifolds whichbelong torelevant matching criteria andthreshold strategy forimage individuals. matching. Wedemonstrate experimentally that theGGCI Image matching isabasic approach tosegmentation approach achieves lowererrorrates andtheGSIM that canbeusedtolocate knownobjects inanimage, to approach brings downthesensitivity ofgrayvalues to search forspecific patterns, etc[2]. Image matching is change inradiometry andreduces multi local extrema to widely applied todetermine stereoscopic sceneproperties someextent. ifmorethanoneimage ofthesame scenetaken from different location isavailable. Inthis paper, we consider