News Video Segmentation and Categorization Techniques for Content-Demand Browsing

The objective of this research is to develop an automatic and efficient news video indexing and management system based on the characteristics of the structure of a complete TV news video in Taiwan. We first employed two features which include the Normalized Color Histogram Intersection of Quantized Hue and Edge Exchanging Ratio to detect the shot boundaries and to segment the corresponding shots within a news video. Next we performed human face detection on the key frames which were extracted from each shot by using harr-like features. After detecting the shots containing the human face, we classified these shots into anchorperson and non-anchorperson shots using “Anchorperson Recognition” procedure. Lastly, according to the programming and structure of a news video, we divided a complete news video into six sections listed as follows: opening animation, anchorperson greeting, headlines preview, news stories, weather forecast and the closing. Experimental results show that our approach can accurately and efficiently segment the news videos into the desired parts mentioned above. As a result, users can quickly browse or retrieve segments of interest from the news videos.