A multi-media database is a type of Database devoted to the storage and the multi-media organization of data: sound documents, images, vidéos. They can be based on various architectures of databases, the most used types being the relational model and the model object.

The databases multi-media are not still applications very current, and still pose problems, in particular for the Indexation of their contents and research by contents.

The traditional approach consists in informing key words describing in a vocabulary restricts the principal and quite identifiable characteristics stored documents. The obvious limitations are the lack of precision of description and impossibility of making research on existing in the documents but nonrelevant information from the point of view of the indexing.

An approach currently prone to research consists in having a dynamic indexing by the application of operators of Image processing (for the databases images and video) or Treatment of the signal (for sound databases). The objective is to break up the documents into elementary entities structured and connected between them, so that the indexing makes it possible to find either of the forms (the structures) or of the objects (entities) or of the combinations of both. The problem then consists in finding “the good” tools for treatment of the signal (in the broad sense) to do the work of analysis of the documents, and building the indices so as to fully exploit extracted information.

The engine of requests is him also a subject of study under these conditions, because the requests are not thus expressed more according to texts and of numerical values “simple” but according to space and temporal relationship between entities which should in addition be described in a structured way.