This proposal presents an innovative approach to improving existing video abstraction techniques in order to automatically obtain a short abstract of a lengthy color video motion picture. The abstract is obtained by examining each frame of the motion picture to detect changes between features of consecutive frames. This improvement is proposed as the first step of video abstraction: feature extraction. The features of each frame are extracted using four techniques: intensity differentiation, edge detection, line detection, and fuzzy medial axial representation. The changes are detected with each of the methods and the results are resolved into a single decision parameter using fuzzy logic for compromising opinions. The advantage of our proposed improvement is that changes are better detected with minimal additional processing cost. The resulting video abstraction will be implemented on a personal computer with a video board to interface with the video source.The video abstraction technique can be used to summarize home videos to help viewers sort the contents of several tapes in a limited time; to organize security video for monitored buildings to better detect unusual activity; to help movie producers get a quick summary of video footages while putting together a movie; and to prepare abstracts of video archives that can be quickly downloaded for browsing over the Internet and the forthcoming National Information Infrastructure.
Keywords: Phase_I, NASA, Abstract, FY94
