Representation of data models via a language specifies essential properties
of the real world described by the data. Some of the most popular languages
(e.g. entity-relationship model, object data modeling language, unified
modeling language) are based around the notions of Abstract
Data Types (ADTs), entities, relationships, constraints, and
rules. Primitive ADTs specify a domain of possible values, a set of operators
and the semantics of operators. Higher level ADTs, e.g., audio, video,
images, have a set of simpler attributes and operations (also called methods).
Entities (e.g. movies, CD-ROMs, journals,
articles, slide-slows, publisher) describe the variety of elements in the
digital library. Entities have special attributes, which uniquely identify
each instance. Entities have relationships,
e.g. a music-video is released by a publisher. Constraints
specify restrictions to ensure data integrity, e.g. whether the
author of an CD-single can be left blank (anonymous).
Current methods for modeling audio, video and image data can be classified
into data-driven and user-driven.
Data driven methods use statistical modeling to fit probability distributions,
where as user-driven methods take advantage of the peculiarities of human
perception. Other data modeling topics relevant to multimedia digital library
include spatial and temporal data models.