Machine learning of event segmentation for news on demand

There are many different types of multimedia videos found in the world today—consider home videos, surveillance camera videos, television broadcasts as general categories. Commercial users, government personnel, and home consumers all have specific requirements to search these videos for topics and/or events. In order to support user query for these elements of interest , multimedia systems must segment and retrieve relevant segments of information. With advances in video digitization, annotation and extraction, automated multimedia processing systems are being created for many of the various video types. In these systems, event segmentation occurs manually, semiautomatically, or automatically. Each type of multimedia video has varying levels of structure. For example, a home video may contain stories of a vacation, child's birthday party, and Christmas morning. The birthday party story may contain events of a child blowing out the candles, opening gifts, and playing games. In some stories , there may only be one event per story. The event pertaining to the child blowing out the candles may contain shots of the child's excitement of the oncoming cake, the friends singing, and the News on Demand FOR OF Deconstructing broadcast news using all sources of input from the multimedia stream.