Požadovaný rozpočet: 1.549.000 EUR

The multimedia databases (MDBs) require specific models and implementation techniques. Despite conventional DBs, the multimedia objects have loosely defined structure and semantics. The query concepts and access methods for MDBs are often based on the notion of similarity, which serves as a ranking function between query and database objects. In this project we focus on four state-of-the-art goals in the area of MDBs: Because the similarity measures are often expensive to compute, there have to be efficient access methods developed, minimizing the CPU costs when querying/indexing a database. The first project goal is a development of efficient access methods allowing a large-scale MDB management supporting metric and non-metric similarities. Second, the current research does consider rather primitive query concepts, like range or kNN queries. In this project we will investigate advanced query types, like re-ranking, metric skylines, metric top-k dominating queries and other multi-queries. Such advanced concepts are necessary nowadays, since the data volumes grow rapidly while the primitive queries lose their expressive power. Third, there exists a gap between the database and domain communities – the domain researchers are often not aware of advanced database techniques. On the other hand, the database researchers often simplify the domain problems just to illustrate the potential impact of their techniques in a particular domain (making just “toy” applications). In this project we aim to disseminate and implant the developments into real domain applications. In particular, we plan to apply the state of-the-art techniques into bioinformatics and medical databases. Finally, we plan to implement the SIRET web engine, a public experimental platform for similarity retrieval of various kinds.

Žadatel: Univerzita Karlova