Retrieval of Sports Video Clips Using Audio-Visual Features and Text Information

Abstract

Video \u00a0clip \u00a0retrieval \u00a0plays \u00a0a \u00a0critical \u00a0role \u00a0in \u00a0the \u00a0content-based \u00a0sports \u00a0video \u00a0retrieval. \u00a0This \u00a0paper proposes a content-based retrieval strategy of sports video clip in which visual and auditory features and text information are extracted to locate similar video clips. Because in sports game play scenes are concerned and interested \u00a0for \u00a0most \u00a0audiences, \u00a0a \u00a0long \u00a0sport \u00a0video \u00a0is \u00a0first \u00a0automatically \u00a0segmented \u00a0into \u00a0play \u00a0segments \u00a0and time-out \u00a0segments \u00a0based \u00a0on \u00a0audio \u00a0feature \u00a0in \u00a0compressed \u00a0domain. \u00a0And \u00a0then \u00a0we \u00a0obtain \u00a0similar \u00a0video \u00a0clips based \u00a0on \u00a0sliding \u00a0shot \u00a0window \u00a0algorithm \u00a0and \u00a0equivalence \u00a0relation \u00a0theory. \u00a0Finally \u00a0similar \u00a0video \u00a0clips \u00a0are ranked \u00a0by \u00a0visual \u00a0factor \u00a0and \u00a0order \u00a0factor \u00a0in \u00a0terms \u00a0of \u00a0longest \u00a0common \u00a0sequence \u00a0(LCS). \u00a0The \u00a0experimental results \u00a0showed \u00a0that \u00a0the \u00a0proposed \u00a0method \u00a0could \u00a0effectively \u00a0and \u00a0efficiently \u00a0retrieve \u00a0similar \u00a0video \u00a0clips corresponding with a query clip in a sports video of TV program.