A Framework for Semantic Video Content Indexing Using Textual Information

In these last years, many works have been published in the video indexing and retrieval field. However, except some specific cases such ass port video where i ts possible t o estimate the set of important events and concepts in the document, this research is generally limited to analyzing low level content. In this paper, we introduce an approach for semantic video indexing that combines two levels of descriptions. First, we extract automatically textuel information from video frames. The second part of our approach consists to exploit linguistic techniques and semantic network in order to extract semantic concepts such as person identity, location name, event type etc. These informations are then used for semantic description of video content. Our proposed approach was tested on video collection of Arabic TV news and experimental results have been satisfying.