Similarity searching in very large multimedia databases

During the last decade, the area of similarity search became very hot topic for both research and commercial applications. Current DB applications use data with considerably less structure and much less precise queries than traditional database systems (e.g. RDBMS). Examples are multimedia data like images or videos that offer query-by-example search, product catalogs with preference-based search, scientific data records from observations or experimental analyses such as biochemical and medical data, and so on. Such data can neither be ordered in a canonical manner nor meaningfully searched by precise database queries that would return exact matches.We focus on development of methods for effective and efficient similarity search in large multimedia databases. The effectiveness tasks deal with relevance of results retrieved from the database based on a query, while we focus on shape extraction and shape matching applied to image databases. The efficiency tasks (a substantial issue of the project) dealwith the performance problems. Due to extremely fast growth of multimedia data, the design of access methods suitable for similarity search cannot be limited just to centralized solutions. Therefore, we focus on distributed indexing, allowing to scale upwith the increasing database sizes. The last objective is a specific design of access methods for non-metric similarity measures, in particular for biological domain and also for local shape matching in image databases.

Keywords

podobnostní hledánídistribuované indexační strukturymetrický prostorextrakce vlastností z dat