Research On A Generalized Multimedia Database Model Based On MPEG-7

With the wide spreading of multimedia technologies in WWW and people's life, the researches on multimedia database management systems (MMDBMSs) have drawn greater attention in recent years. Multimedia data have many types, complex structures and rich semantics. Most existing MMDBMSs lack unified management of a variety of media types and can not provide effective semantic retrieval. Hence, it is significant to design a generalized multimedia database model to support most media types, apply to most media applications, and study the storage, management and retrieval methods of multimedia semantics.Focusing on the research on a generalized multimedia database model, the studies of its storage strategy and semantic retrieval method especially the intelligent retrieval of high- level semantics are carried out, and a prototype system is implemented.After summarizing the requirements of generalized multimedia database models and analyzing the advantages and disadvantages of the Multimedia Content Description Inter- face (MPEG-7), an MPEG-7 based generalized multimedia database model called G3M is proposed and formally defined, which supports a variety of media types (e.g., audio, video, images, graphics, text and cross-media data, etc.). Existing MPEG-7 based Multimedia Database (MMDB) solutions have two main problems: one is that the valuable metadata have been neglected, the other is that none solves the nature defect of MPEG-7 that it has no formal grounding. G3M analyses the MPEG-7 metadata throughout, and uses multi- level abstraction to describe different aspects of multimedia, including normal attributes, segmentation, low-level features, spatio-temporal semantics and high-level semantics. Knowledge ontologies are integrated with G3M schemas and instances seamlessly to solve the no formal grounding problem of MPEG-7 and enhance the semantic description ability of G3M. The generalization of G3M is proved by analyzing how G3M satisfies the requirements of generalized multimedia database models. With the help of imported domain knowledge, G3M can better serve for special applications.To effectively support the logic model of G3M, a dynamic and hybrid storage strategy is proposed after comparing the existing XML database solutions and MPEG-7 based MMDB storage methods. A multi-grained storage method of G3M is dynamically designed according to user preferences or application requirements. Different data types in MPEG-7 are analyzed carefully and given appropriate storage methods in G3M. A relational schema is defined for multimedia ontologies, also using the hybrid storage strategy, which is good for ontology-based query reasoning. Experiment results prove that G3M is superior to the other two storage solutions, a pure XML storage solution on DB2 and a representative relational shredding method called SM3+, in semantic pruning query, composite query and scalability performance.MPEG-7 based XQuery and ontology-based semantic query methods are designed for G3M. Users may not know the exact definition of G3M, but query G3M with MPEG-7 based XQuery, which will be automatically turned to the equivalent query statements on G3M schemas. It ensures the interoperability between G3M and other MPEG-7 based multimedia systems. Ontology-based semantic queries in G3M are built on classic query engine of a relational DBMS. To enhance the semantic representation and query capabili- ties of G3M, knowledge concepts, attributes, relationships and rules are applied to the high-level semantic abstraction and implicit semantic extraction on G3M with effective query rewriting methods transparent to users.A prototype system is designed and built on IBM DB2 V9, while a G3M model generator, a hybrid parser, an ontology-based query rewriting engine and a personalization processing unit are added to simulate the G3M server. The front side is a user-centric personalized multimedia modeling and retrieval tool, which supports user preference setting, background knowledge importing, dynamic modeling of G3M and personalized semantic query based on ontologies. All G3M technologies mentioned above are united in this prototype system, which provides a unified experiment platform for our generalized multimedia database model modeling and semantic query research.