Chapter4Featuredetectionandmatching4.1Pointsandpatches................................1834.1.1Featuredetectors............................1854.1.2Featuredescriptors...........................1964.1.3Featurematching............................2004.1.4Featuretracking............................2074.1.5Application:Performance-drivenanimation..............2094.2Edges......................................2104.2.1Edgedetection.............................2104.2.2Edgelinking..............................2154.2.3Application:Edgeeditingandenhancement..............2194.3Lines......................................2204.3.1Successiveapproximation.......................2204.3.2Houghtransforms............................2214.3.3Vanishingpoints............................2244.3.4Application:Rectangledetection....................2264.4Additionalreading...............................2274.5Exercises....................................228R.Szeliski,ComputerVision:AlgorithmsandApplications,TextsinComputerScience,181DOI10.1007/978-1-84882-935-0_4,©Springer-VerlagLondonLimited20111824Featuredetectionandmatching(a)(b)(c)(d)Figure4.1Avarietyoffeaturedetectorsanddescriptorscanbeusedtoanalyze,describeandmatchimages:(a)point-likeinterestoperators(Brown,Szeliski,andWinder2005)c2005IEEE;(b)region-likeinterestoperators(Matas,Chum,Urbanetal.2004)c2004Elsevier;(c)edges(ElderandGoldberg2001)c2001IEEE;(d)straightlines(Sinha,Steedly,Szeliskietal.2008)c2008ACM.4.1Pointsandpatches183Featuredetectionandmatchingareanessentialcomponentofmanycomputervisionappli-cations.ConsiderthetwopairsofimagesshowninFigure4.2.Forthefirstpair,wemaywishtoalignthetwoimagessothattheycanbeseamlesslystitchedintoacompositemosaic(Chapter9).Forthesecondpair,wemaywishtoestablishadensesetofcorrespondencessothata3Dmodelcanbeconstructedoranin-betweenviewcanbegenerated(Chapter11).Ineithercase,whatkindsoffeaturesshouldyoudetectandthenmatchinordertoestablishsuchanalignmentorsetofcorresp