Content-Based Analysis Of Video Structure

With the development of information technology, digital video has become the main media form in information systems. How to effectively manage and analyze the massive videos becomes an important research topic. Video structure analysis is the basis of analyzing videos.Digital video is a pyramid-like structure, from bottom to top, there are frames, shots, scenes and whole video. The video structure analysis includes shot boundary detection, key frame extraction and scene segmentation. In this thesis, I introduce scale invariant feature transform into video structure analysis area. Experimental results show that the method is effective. The main content of this thesis is showed as below.1. Shot boundary detection.a) Give the definition, classification and some samples of video shot boundary.b) Study the existing shot boundary detection technologies.c) Propose a new method to detect shot boundary using scale invariant feature transform. This method can detect cut and gradual shot boundary at once.d) Experimental results show our method is effective in detecting both type of shot boundary:cut and gradual, besides, it's robust to shot rotate, scale, noise and lustrous shine.2. Key frame extraction.a) Give the definition of key frame and principle of extracting key frame.b) Study the existing key frame extraction technologies.c) Propose a new method to extract key frames. Firstly, the video is segment into video clips according to content coherent, then the frame with most SIFT key point in each clip is extracted as the key frame.d) Experimental results show that the extracted key frames are representative and low-redundancy in vision content.3. Scene segmentation.a) Give the scheme of segment video into scenes.b) Study the key problems of Scene segmentation.c) Propose a Scene boundary detection algorithm. I use the number of matched SIFT key point of two shots'key frames to calculate the similarity of the shots. Then I choose the time-adaptive grouping method to cluster shots.d) Experimental results show SIFT is an effective feature to segment video into scenes.4. Video structure analysis systema) Propose a scheme of video structure analysis system based on scale invariant feature transform.b) Design a video structure analysis system. The main functions of the system include:SIFT key point extracting and matching, shot boundary detection, key frame extraction, scene segmentation and so on.