News story segmentation based on audio-visual features fusion

This paper presents a method for news video story segmentation, which fuses multi-feature including audio and visual. At first, this paper detects the anchorperson shot for news video and determines the beginning of news story, and then detects topic caption between anchorperson shots. In the next step, silence clips in news video are detected using short-time energy and short-time average zero-crossing rate parameters, and then voice features of anchorperson is analyzed. At last, this method fuses multi-feature such as anchorperson shot, topic caption, silence and voice feature to segment news stories. Experimental results show that the approach is valid and avoid the deficiency of detecting news story by a single feature.

I. Introduction

Video interpretation or video understanding is one of the most exciting research areas in recent years.^[1]–[3] In general, users remember video contents in terms of events or stories. Thus there is a need to organize video contents in terms of small, logic units that represent the conceptual chunks in users' memory. Such unit is called story. News video is a specific kind of video. The segmentation of news video into story units is an important step towards effective processing and management of large news video archives. However, story segmentation is not an easy job, because video data is essentially combination of texts, video, sound, images and other media. These media are not isolated, but have semantic associations between each other. Therefore, we analyze comprehensive semantic information which contained in the video data in order to get the video clip complying with the people's requirements.