Content-based retrieval in multimedia databases

Publisher Website
Google Scholar
Abstract
The flexible retrieval of pictorial data on the basis of content hinges on the ability to model multimedia data effectively. In using the entity-attribute-relationship model, the main semantic concepts of entity, attribute and relationship can be represented in a simple manner which correspond very closely to the noun, adjective and verb which are the essential components of a simple description of a picture. Such a system has an advantage over traditional ways of representing the data in an automated system in that it allows the concept of relationships between two (or more) objects within a picture to be represented in the database.Our basic starting point is man-machine cooperation in which the pattern processing capability of the human eye is fully exploited. As a result, it should provide a much better degree of flexibility in the identification of complex high-level patterns. We also introduce a picture description language which allows the contents of pictures to be systematically represented, these resultant representations can then be stored for subsequent processing. Picture retrieval is accomplished by specifying picture queries using a picture query language. Evaluation experiments carried out on a prototype system show that this approach is able to yield precision and recall performances of over 62% and 88% respectively.

This publication has 9 references indexed in Scilit: