On-line video skimming based on histogram similarity

Abstract

This paper describes the method proposed for the TRECVID 2007 BBC rushes summarization task. Such method has been developed starting from an on-line summary generation approach which provides a fast way to generate a base summary that can be later reduced with an off-line summarization technique if the target summary size is not reached. The summarization algorithms aim to reduce the visual redundancy of the video by methods based on color similarity features and targeting to the generation of connected video skims (instead of providing isolated keyframes). The on-line summarization process provides a very fast method with low memory usage for video summary creation.