News Video Scene Segmentation Research

With the development of modern computer technology, especially data storage and transmission, digital video has been widely used in the fields of news, education, entertainment and medical service etc.. On the one hand, video embraces a number of advantage, e. g. vividness, powerful expressive force and enormous information capacity; on the other hand, the complexity of the data structure and the huge magnitude of the data capacity as well as the least transparence of the video information bring much inconvenience for video information organization, browse and retrieval. The inconvenience has badly embarrassed the application of video. The motivation of video parsing techniques is to make the video media more structured, it is the basis and premise of many techniques including content-based retrieval, video nonlinear editing and browsing. Hence it plays an important role in video research domain. TV news is one type of video. As an important program type, it is a key source for each family to timely know what happened in the world. The dissertation addresses the scene segmentation and retrieval techniques of news video. The research focuses on many aspects, including: shot detection; anchorperson shot detection, close-caption detection, scene segmentation and retrieval techniques of news video. Some impressive contributions are listed as follow.1. An algorithm for shot detection of different flashlights combination is proposed. It detectsthe flashlights combination exactly and overcomes the weakness brought by the presentand absent of combination, in terms of differences between histograms of flames.Compared with other methods, this algorithm adapts flashlights better, which provides ahigher precision. 2.A new method for anchorperson shot detection is presented based on histograms of primary colors in half screen, where detection is in accordance with the frequencies of anchorperson shot in the news. Particularly, the screen is departed into two parts which are detected separately, removing flaws arising from the changes in background and small windows. The method aforementioned implements a complete auto-detection in a faster speed acquiring a higher precision and it will find its places in almost every type of news.3.A new method for detecting close-caption in this paper uses both the shape and location of characters to get exact results in a fast speed with the, principle and speeding-up mechanism discussed in detail. This method breaks through the deep dependence upon manual work and weak adaptability .It may be applied to the auto-establishment of catalogue and news expression, which makes comprehension of news contents after scanning a few images possible. Fortunately, we get a satisfied prevision and recall. Due to its robustness, it may be applied to various news 4.Automated segmentation of news scene based on significative tissues is depicted. By analyzing of the presentation of anchorperson shots, close-caption and mute section in change points, the drawbacks of weak adaptability in exited methods is made up in segmentation.The stand of this paper is a complete automated detection along with high precision and flexibility for various news video, keeping a fast speed as possible. Courting the goal of theorganization and retrieval of vast video data, we study news video resolution and scene segmentation in depth and propose several effective methods for detection. In addition, some heuristic trials on retrieval of news video based on character information in close-caption are made and its feasibility is proved too.