Color superhistograms for video representation

In this paper we present a novel method for computing superhistograms to represent video segments. We build cumulative histograms for video shots and scenes. The underlying assumption is that a TV program has a consistent color palette, which can be derived as a family of merged individual shot histograms. We present the experimental results computed from several hours of different TV programs such as sitcoms, serials and news. The superhistograms representing episodes of the same sitcom look strikingly similar. The histograms extracted from TV news are not similar to the histograms extracted from sitcoms. However, there are superhistograms extracted from different sitcoms, which show certain resemblance to each other. This method can be used in a variety of applications that need video classification and retrieval methods such as video editing, studio archival, digital libraries, consumer products, and web crawling.