Video description by automatic content extraction

The richness of semantic content of visual representation is related to the fact that it has an aspect right only for interpretation by human. When videos are stored in a database their retrieval is normally performed simply by name. The need to have information on their content, during the retrieval phase, is growing and its extraction constitutes a challenge if performed automatically by the retrieval system itself. Various researchers attempted to associate each film to a global description that is automatically extracted and is a way to index it in a database. The Moving Picture Expert Group actually is working to a new standard, named MPEG-7, which deals with a multimedia content description interface, that does not include an automatic extraction of descriptions and says nothing about a browser for showing these descriptions. The work presented focuses on two aspects: extraction of a compact representation of a frame series and of a description based on little number of features; and definition of a query language constituted by a set of predicates and facts; these last are derived by a series of features automatically extracted from the frame sequence. A main tool to extract a compact representation from a series of frames is mosaicing.