TV News Database Indexing System with Video Structure Analysis, Representative Images Extractions and OCR for News Titles

In this paper, we present the semiautomatic system for analysis of the temporal structure of a news show, where the indexing of TV news program material is done following a classification resulting from the segmentation of video sequences, the definition of areas of interest and the recognition of news titles through automatic optical text recognition. Several experiments are presented regarding video segmentation methods, text extraction methods, and models of temporal and spatial structures that define a TV news program.