Personal television: A crossmodal analysis approach

Abstract

Personal information consumption became more and more important due to the huge number of existing information channels and the broad range of available information. While obtaining information from the Internet, it is usual to create individual profiles in order to consume personalized content based on the users preferences, e.g. within news portals. The traditional broadcast domain does not offer such functionality by default. But there are still a number of scenarios for personal TV imaginable. This paper describes two of such scenarios, dealing with cross-modal audio-visual analysis methods. Furthermore, a system based on news segmentation and a news clip module is presented. The news segmentation algorithm uses a knowledge-based approach and decision rules, whereas the annotation algorithm focuses on text capturing components e.g. optical character recognition and speech recognition. Finally, the results of an in-depth evaluation based on German newscasts in TV are presented.