简介：Westudythecombinationofsymbolfrequenceanalysisandnegativeselectionforanomalydetectionofdiscretesequenceswhereconventionalnegativeselectionalgorithmsarenotpracticalduetodatasparsity.TheoreticalanalysisonergodicMarkovchainsisusedtooutlinethepropertiesofthepresentedanomalydetectionalgorithmandtopredicttheprobabilityofsuccessfuldetection.Simulationsareusedtoevaluatethedetectionsensitivityandtheresolutionoftheanalysisonbothgeneratedartificialdataandreal-worldlanguagedataincludingtheEnglishWikipedia.Simulationresultsonlargereferencecorporaareusedtostudytheeffectsoftheassumptionsmadeinthetheoreticalmodelincomparisontoreal-worlddata.

简介：

简介：Inthispaper,wepresentanovelapproachtosynthesizingfrontalandsemi-frontalcartoon-likefacialcaricaturesfromanimage.Thecaricatureisgeneratedbywarpingtheinputfacefromtheoriginalfeaturepointstothecorrespondingexaggeratedfeaturepoints.A3Dmeanfacemodelisincorporatedtofacilitatefacetocaricaturesbyinferringthedepthof3Dfeaturepointsandthespatialtransformation.Thenthe3Dfaceisdeformedbyusingnon-negativematrixfactorizationandprojectedbacktoimageplaneforfuturewarping.Toefficientlysolvethenonlinearspatialtransformation,weproposeanovelinitializationschemetosetupLevenberg-Marquardtoptimization.Accordingtothespatialtransformation,exaggerationisappliedtothemostsalientfeaturesbyexaggeratingtheirnormalizeddifferencefromthemean.Non-photorealisticrendering(NPR)basedstylizationcompletesthecartooncaricature.Experimentsdemonstratethatourmethodoutperformsexistingmethodsintermsofviewanglesandaestheticvisualquality.