With the rapid growth of multimedia data, a lot of attention has been recently devoted to the development of multimedia retrieval systems. The research has followed two main directions: The first one applies existing text-search mechanisms to retrieve multimedia data based on its descriptive annotations, the second approach retrieves data by content. In case of text-based searching, the quality of results depends on the quality of text metadata, which is often not very high (especially in large general-purpose collections such as web image galleries). In the content-based approach, data objects are indexed and searched using features extracted from the data that describe their important characteristics. However, this solution suffers from the well-known semantic gap problem, i.e. the discrepancy between the similarity as computed using the descriptors and human understanding of similarity.