简介:Invideoinformationretrieval,keyframeextractionhasbeenrecognizedasoneoftheimportantresearchissues.Althoughmuchprogresshasbeenmade,theexistingapproachesareeithercomputationallyexpensiveorineffectiveincapturingsalientvisualcontent.Inthispaper,wefirstdiscusstheimportanceofkeyframeextractionandthenbrieflyreviewandevaluatetheexistingapproaches.Toovercometheshortcomingeoftheexistingapproaches,weintroduceanewalgorithmforkeyframeextractionbasedonunsupervisedclustring.Meanwhile,weprovideafeedbackchaintoadjustthegranularityoftheextractionresult.Theproposedalgorithmisbothcomputationallysimpleandabletocapturethevisualcontent.Theefficiencyandeffectivenessarevalidatedbylargeamountofreal-worldvideos.