| With the rapid development of multimedia, network technique, wide application of science and technology and improvement of people's life, there are more and more chances for people to touch with video information. How to effectively organize and manage the new type of media carrying enormous volume of information, which can help us quickly get the video in need exactly, have become an imperative and a heated subject in multimedia field. Based on the above actual demand, Content-Based Video Retrieval technology has aroused widely concern. Content based video retrieval searches object through shot detection, key-frame selection and retrieval techniques. The result won't be disturbed different comprehension for different videos.Against this background, this dissertation studies the key technology of compressed domain video retrieval, including shot segmentation, key frame selection and video retrieval. The main contents are as follows:1) Common shot converting forms and shot segmentation algorithms in the extract domain are discussed. Based on analyzing the actuality of shot detection in the extract domain and H.264 standard, this dissertation has proposed rapid methods for shot detection, which are based on the information derived from the partially-decoded H.264 video bit stream. Finally, the effectiveness of this algorithm we proposed is verified through comparing with other algorithms.2) Some common key frame selection algorithms in extract domain are discussed and the principle of key frame selection is also introduced. This dissertation improves a fast and effective key frame selection algorithm to adapt better and without significantly increasing the calculation. It is very flexible in deciding the key frame numbers.3) A hierarchical video retrieval algorithm is proposed in this dissertation. Hierarchical structure is used in the video management and analysis. A feedback mechanism is introduced to make the system return batter results when using the system mquiry. |
PDF Full Text Request