The Research On Digital Video Shot Detection

Shot boundary detection is always an important topic in digital video processing.Shots are the most basic elements which compose the video sequence. Shot detectionprovides the foundation for video structure parsing and high semantic levelprocessing.In this thesis, the previous work done in shot detection is introduced. An abrupt shottransition detection method based on the statistical model and a gradual shot transitiondetection method based on the EM curving fitting are proposed in order to resolve theproblems in current methods.The following is the main contents and contributions of this thesis: Basic concepts, research background and research meanings are introduced. The traditional algorithms are systemically analyzed, and the issues which are focused are proposed. The issues of shot detection (similarity measurement, motion detection, illumination change processing) are analyzed in detail. In similarity measurement part, intensity, histogram, motion feature and edge based feature are introduced; In motion detection part, the method avoiding false detection caused by camera and object motion are introduced; In illumination change processing part, the common method processing illumination change are introduced. An abrupt shot transition detection method based on statistical model is proposed in order to resolve some issues. In this method, a motion compensation feature are adopted to decrease the false detection caused by motion; the statistical model are used to model the abrupt transition, the pre-probability and post-probability are considered; finally, the flash light are processed to avoid false detection. In order to speed up the method, a skip frame strategy is adopted. A gradual transition detection method based on EM curve fitting is proposed. First the adaptive threshold is used to select the candidate gradual transition in a sliding window; Second Kalman filter is adopted to search the exact boundary of gradual transition; Third EM curving fitting are used to fit the gradual transition curve and get a series of parameters; Finally, the decision tree is adopted to decide whether there is a gradual transition.