A Semantic and Personalized Framework for News Video Retrieval Based on Textual and Visual Transcripts

Abstract

Multimedia data have recently been available in several sources (Web, specialized corpora, etc.). The importance of TV news video comes particularly from the fact that they inform viewers about the situation of other peoples in a given place of the world and in a given period of time. The real problem is how user will access quickly and efficiently in large scale collection of news videos. In this paper, we present a compound approach integrating a semantic multi-modal analysis of video data in order to explore such content. Firstly, the summarizing process whose goal is to accelerate the video content browsing based on genetic algorithms. Secondly, the indexing process, which operates on video summaries, is based on text and image. Thirdly visualization maps are, then, generated to optimize the browsing video collection. It can also be considered as a knowledge discovery tool because we assist users in exploring large video corpora with no required deep interaction with the offered tools. Evaluations were conducted in the TRECVID framework described in the two ultimate sessions.