EXTENSIBLE OBJECT MANAGER FOR MULTIMEDIA DATABASE MANAGEMENT SYSTEMS

Abstract

Multimedia objects (such as video, image, voice, graphics) have imposed new requirements and research considerations on the design of database management systems. The presence of spatio-temporal and composition relationships among them, their large volume and their inherent distribution pose interesting modeling and implementation requirements. As multimedia database management systems integrate a variety of data types they require an extensible architecture to allow application designers to define new types and should account for the evolving nature of the hardware and software support for such multimedia objects. This thesis presents the design and implementation of an object manager for multimedia database management systems, called Extensible Multimedia Object Manager (EMOM). EMOM is a module in a previously defined multimedia database architecture at the Institute of Systems Science. The primary goal in designing EMOM is to provide extensibility, the ability to dynamically add new data types and new system modules to support the evolving multimedia objects. EMOM addresses extensibility by providing an extensible object identity (OID) mechanism, self descriptive object model and a Java class loader. OID mechanism captures media semantics, class hierarchy and multiple representations of multimedia objects. An object-oriented database, SHORE, is used as the storage manager for storing multimedia objects. Java language is used as the modeling language for defining the characteristics of multimedia objects. Our object model highlights a new method of organizing multimedia objects. In this model, inference methods which are Java classes for processing its own data are embedded within the object. EMOM retrieves an object from the storage manager by its OID. While loading the object from the storage manager, Java classloader of EMOM interprets the Java classes within the data instantiates and executes them. As each object has its own processing or inference methods, new object types are supported dynamically, making the system extensible.