This thesis has investigated data management in multimedia digital libraries, and its implications on the design of database management systems. The main problem of multimedia data management is providing access to the stored objects. The content structure of administrative data is easily represented in alphanumeric values. Thus, database technology has primarily focused on handling the objects' logical structure. In the case of multimedia data, representation of content is far from trivial though, and not supported by current database management systems.
The information retrieval (IR) community has since long studied the retrieval of text documents by their content. Also, research in topics like computer vision and image analysis has led to content-based retrieval techniques for querying image and audio collections. Retrieval systems based on these ideas are typically standalone systems that have been developed for very specific applications. There is not much consensus on the integration of these techniques in general-purpose DBMSs. State-of-the-art solutions simply make new functions available in the query language. These functions interface to otherwise still standalone software systems. This leaves to the user the burdens of both query formulation and the combination of results for each single representation into a final judgement. Also, this leads to inefficient query processing for queries involving several content representations.
Like any DBMS, a MM-DBMS is a general-purpose software system that supports various applications; but, the support is targeted to applications in the specific domain of digital libraries. Four new requirements have been identified for this domain: (1) multimedia objects can be active objects, (2) querying is an interaction process, (3) query processing uses multiple representations, and (4) query formulation provides content independence. The Mirror Architecture and its implementation in the Mirror DBMS therefore provide basic functionality for the management of both the content structure and the logical structure of multimedia objects.
In the Mirror DBMS, content management and databases are completely integrated. Recognizing the strong relationship with IR query processing, the inference network retrieval model has been adapted for multimedia retrieval. The logical algebra of the DBMS has been extended with operators for probabilistic inference in this retrieval model. This approach to integration enables the study of new query optimization techniques, and simplifies the introduction of parallellism and distribution in IR query evaluation.
Other characteristics of multimedia digital libraries demand the support for distribution of both data and operations, and extensibility of data types and operations. As a solution, the integration of advanced middleware and database technology is proposed to replace the monolithic design of traditional database systems. Again, this idea has been worked out in a prototype implementation.
The proposed MM-DBMS architecture has been evaluated in three ways, using the Mirror DBMS prototype implementation. First, the advantages of the integration of content management in the Mirror DBMS have been illustrated by several example queries capturing different information needs. The multimedia IR model developed in this thesis has been tested in some small-scale experiments in the domains of music and image retrieval, confirming that reasoning with multiple representations is both possible and useful. Finally, the execution performance of IR query processing has been evaluated using a standard text retrieval benchmark.