Digitrecognitionfromanaturalscenetextinvideosurveillance/broadcastingapplicationsisachallengingresearchtaskduetoblurred,fontvariations,twisted,andnon-uniformcolordistributionissueswithadigitinanaturalscenetoberecognized.Inthispaper,tosolvethedigitnumberrecognitionproblem,aprincipal-axisbasedtopologycontourdescriptorwithsupportvectormachine(SVM)classificationisproposed.Thecontributionsofthispaperinclude:a)alocaldescriptorwithSVMclassificationfordigitrecognition,b)higheraccuracythanthestate-of-theartmethods,andc)lowcomputationalpower(0.03second/digitrecognition),whichmakethismethodadoptabletoreal-timeapplications.
