Chapter15ConclusionInthisbook,wehavecoveredabroadrangeofcomputervisiontopics.Startingwithimageformation,wehaveseenhowimagescanbepre-processedtoremovenoiseorblur,segmentedintoregions,orconvertedintofeaturedescriptors.Multipleimagescanbematchedandregistered,withtheresultsusedtoestimatemotion,trackpeople,reconstruct3Dmodels,ormergeimagesintomoreattractiveandinterestingcompositesandrenderings.Imagescanalsobeanalyzedtoproducesemanticdescriptionsoftheircontent.However,thegapbetweencomputerandhumanperformanceinthisareaisstilllargeandislikelytoremainsoformanyyears.Ourstudyhasalsoexposedustoawiderangeofmathematicaltechniques.Theseincludecontinuousmathematics,suchassignalprocessing,variationalapproaches,three-dimensionalandprojectivegeometry,linearalgebra,andleastsquares.Wehavealsostudiedtopicsindiscretemathematicsandcomputerscience,suchasgraphalgorithms,combinatorialopti-mization,andevendatabasetechniquesforinformationretrieval.Sincemanyproblemsincomputervisionareinverseproblemsthatinvolveestimatingunknownquantitiesfromnoisyinputdata,wehavealsolookedatBayesianstatisticalinferencetechniques,aswellasma-chinelearningtechniquestolearnprobabilisticmodelsfromlargeamountsoftrainingdata.AstheavailabilityofpartiallylabeledvisualimageryontheInternetcontinuestoincreaseexponentially,thislatterapproachwillcontinuetohaveamajorimpactonourfield.Youmayask:whyisourfieldsobroadandaren’tthereanyunifyingprinciplesthatcanbeusedtosimplifyourstudy?Partoftheanswerliesintheexpansivedefinitionofcom-putervision,whichistheanalysisofimagesandvideo,aswellastheincrediblecomplexityinherentintheformationofvisualimagery.Insomeways,ourfieldisascomplexasthestudyofautomotiveengineering,whichrequiresanunderstandingofinternalcombustion,mechanics,aerodynamics,ergonomics,electricalcircuitry,andcontrolsystems,amongothertopics.Computervisionsimilarlydrawsonawidevarietyofsub-disciplines,whichmakesitchallengingtocoverinaone-semestercourse,letalonetoachievemasteryduringacourseofgraduatestudies.