ABSTRACT
Video adaptation can be classifi ed in many diff erent ways depending on the application scenario used. Some appropriate classifi cations include the following [2]:
Semantic-level video adaptation Signal-level video adaptation (transcoding) Adaptation of scalable streams
Semantic video adaptation can be basically described as the detection of important or relevant fragments of a video clip (like goals in a soccer video or dialog scenes in an interview) and giving higher priority to these segments during the reduction of the resources allocated to the adapted video. Semantic video adaptation of sports videos has been studied extensively in the literature. For instance, in [3], metadata is combined with video analysis to detect important events and players. In [4], nonimportant video segments are replaced with still images, audio only, or text only representations, resulting in signifi cant reduction in the resources required for the consumption (transmission, decoding, and displaying) of a video clip. In [4], an experiment performed with baseball video clips demonstrated that the nonimportant segments occupied more than 50% of the video clip.
