A new support for objects classification in Multimedia Information Retrieval

Multimedia databases store huge amount of heterogeneous information, but the user’s queries usually search for just very short sections of data hidden and mixed with each other. This work presents a support methodology for Information Retrieval Systems on a collection of Multimedia Data Objects. The main idea of this retrieval methodology exploits raw metadata information stored in multimedia objects to realize a classification using an innovative approach, based on a spatial dispersion index. A convenient synthetic representation of multimedia objects is drawn from the Lexical Database WordNet. It provides the system with synonymic and polysemic Semantic Knowledge. With the aim to achieve an alternative segmentation on document classes, a clustering algorithm based on the Nearest Neighbour geospatial index is finally used.