HETEROGENEOUS MULTIMEDIA REASONING

Although various multimedia systems exist in the commercial market today, most development efforts have proceeded along ad hoc lines that apply in narrow domains. The authors propose central principles around which multimedia database systems can be uniformly built for a variety of applications, independent of domain. These principles can be implemented in a shell/compiler. Domain-specific components, which vary from application to application, can be added to this compiler. The authors show that for most media data, it is possible to build meta data atop the physical medium. This meta data can be generated by a human, an image processing program, or a combination of the two. The data may also involve various degrees of certainty and reliability. A unified query language can be used to query this meta data. The language has a well-understood syntax and semantics rooted in formal logic. Based on this formal model, the authors have developed a prototype implementation of a system called the MACS (Media Abstraction Creation System), which provides the shell facilities alluded to above. In parallel, they have devel oped a general-purpose framework called Hermes (Heterogeneous Reasoning and Mediator System), which uses the concept of a mediator to integrate diverse forms of information. MACS and Hermes can be combined to provide efficient support for multimedia reasoning with access to heterogeneous data sources.