A sequence analysis system for video databases

Publisher Summary The proliferation of digital transmissions and video services will lead to a demand for efficient and flexible local storage devices for time-shifting, personal archival, and fast access to downloaded material. Visual search tools must be provided for easily locating specific information within huge volumes of video data. This chapter presents an automatic system for abstracting visual content from compressed video sequences. This has been achieved by identifying a reliable parsing technique, defining a suitable measure for representation of content, and by introducing a new approach for key-frame allocation. Through this method it is possible to control the number of key-frames extracted from a video sequence. The extraction is not based on any parameter setting, but is fully automated. The applied numerical algorithm optimally delivers locations for the assigned amount of key-frames in each shot based on a suitable measure of the activity of the sequence. The analysis of the content gives support for a high representativity of the selected key-frames.