Automatic Segmentation of News Items Based on Video and Audio Features.pdfVIP

Automatic Segmentation of News Items Based on Video and Audio Features Weiqiang Wang Wen Gao Institute of Computing Technology, Chinese Academy of Sciences, Beijing , China, 100080 Email: {wqwang, wgao}@ Abstract. In the paper, we present an approach that exploits audio and video features to automatically segment news items. Integration of audio and visual analysis can overcome the weakness of the approach only using the image analysis techniques. It brings our approach with more adaptation to variable existence situations of news items. The proposed approach identifies silence segments in accompanying audio, and integrates with shot segmentation results, as well as anchor shot detection results, to determine boundaries between news items. Experiments show that the integration of audio and video features is effective to solve the problem of automatic segmentation of news items. 1. Introduction Most early studies in video structure analysis are based on visual information. To effectively index and retrieve video documents, they are segmented into scenes[1,2]. Furthermore, Shot boundaries are determined to characterize content details[3,4], and key frames are extracted to construct index[5,6]. Audio, as another time-dependent media in video documents, can supplement visual information, and supply a unique cue for video content analysis. For instance, in anchor shots of CCTV news, visual content is almost unchanged, but it is possible multiple news items are reported by anchorpersons synchronously. In a movie, a group of consecutive shots maybe differs drastically in visual content, but the accompanying music indicates they belong to the same semantic clip. Recently, more literatures proposed to apply audio analysis techniques in characterizing video content. [7] exploited multiple audio features and a neural net classifier to differentiate five classes of TV programs, including advertisement, basketball, football, news, weather. [8] pr