Inthispaper,afuzzyreasoningbasedtemporalerrorconcealmentmethodisproposed.ThebasictemporalerrorconcealmentisimplementedbyestimatingMotionVector(MV)ofthelostMacroBlock(MB)fromitsneighboringMVs.WhichMVisthemostproperoneisevaluatedbysomecriteria.Generally,twocriteriaarewidelyused,namelySideMatchDistortion(SMD)andSumofAbsoluteDifference(SAD)ofcorrespondingMV.However,eachcriterioncouldonlypartlydescribethestatusoflostblock.Toaccomplishthejudgementmoreaccurately,thetwomeasuresareconsideredtogether.ThusarefinedmeasurebasedonfuzzyreasoningisadoptedtobalancetheeffectsofSMDandSAD.TermsSMDandSADareregardedasfuzzyinputandtheterm‘similarity’asoutputtocompletefuzzyreasoning.ResultoffuzzyreasoningrepresentshowthetestedMVissimilartotheoriginalone.Andk-meansclusteringtechniqueisperformedtodefinethemembershipfunctionofinputfuzzysetsadaptively.Accordingtotheexperimentalresults,theconcealmentbasedonnewmeasureachievesbetterperformance.
