Research And Application Of Feature Based Video Mining Technology

With the development of video mining application, it is necessary to do a deep research on video mining technology. Video data, which contain various video features and enormous data obtained from video features, are not only the foundation of video database, but also the necessary conditions of video mining. Therefore, feature based video mining technology is an important work of research on video mining technology.This paper analyses key technique and researches on two essential parts of feature based video mining. The first part is shot boundary detection with rapid motion, and the second is multi-modal fusion based video abstraction.There are many kinds of video in our life, including video with rapid motion. However, shot is the basic unit of video sequence. And there are two different shot changes, abrupt change and gradual change. Since that, a rapid motion tolerated shot boundary detection method is required. In this paper, an integrated abrupt shot boundary detection method of two-threshold twice difference method and filter mechanism is presented. The method makes use of validity of twice difference method and flexibility of filter mechanism. Based on this method, a rapid motion tolerated gradual shot boundary detection method is advanced. Contrasting the results of these experiments to classical methods', these methods eliminate the negative influence of shot detection's accuracy arose by rapid motion. And these methods accord with the requirement of shot boundary detection on video containing rapid motion.Video abstraction is a short summary to a longer video. It can present significant content of video briefly. It is intent to locate important information in video fast and conveniently. It can effectively solve the problem which called information explosion. Since video contains not one-form information, but several-forms like video information and audio information, one-modal video abstraction can not fit. Thus, this paper applies multi-modal fusion method to video abstraction. It is used to extract video information, audio information and texture information from video adequately. By the results of these experiments, multi-modal fusion based video abstraction is better than one-modal.