Retrieval by content of commercials based on dynamics of color flows

The huge amount of data associated with videos complicates the extraction of salient information content descriptors from low level video features. Commercials are a video category where a large part of the content depends on low level perceptual features such as colors and color dynamics. These are related to changes-in terms of shrinking, growth and translation-of colored regions over consecutive frames. By changing its shape and position over consecutive frames, each region defines a 3D volume: a color flow. In this paper, a system is presented that allows retrieval of commercials based on color flow similarity. The system relies on automatic extraction of color flows, and description of color flows based on 3D wavelet decomposition