摘要
Invideoinformationretrieval,keyframeextractionhasbeenrecognizedasoneoftheimportantresearchissues.Althoughmuchprogresshasbeenmade,theexistingapproachesareeithercomputationallyexpensiveorineffectiveincapturingsalientvisualcontent.Inthispaper,wefirstdiscusstheimportanceofkeyframeextractionandthenbrieflyreviewandevaluatetheexistingapproaches.Toovercometheshortcomingeoftheexistingapproaches,weintroduceanewalgorithmforkeyframeextractionbasedonunsupervisedclustring.Meanwhile,weprovideafeedbackchaintoadjustthegranularityoftheextractionresult.Theproposedalgorithmisbothcomputationallysimpleandabletocapturethevisualcontent.Theefficiencyandeffectivenessarevalidatedbylargeamountofreal-worldvideos.
出版日期
1999年03月13日（中国期刊网平台首次上网日期，不代表论文的发表时间）
