Media processing and retrieval model for multimedia documents databases.

Abstract

Typically, multimedia applications involve thousands of hours of video, images, audio, text and graphics that need to be stored, retrieved and manipulated in a large multimedia database. There is therefore an important need for novel techniques and systems which provide an efficient retrieval facility of the voluminous information stored in the multimedia database. Such a facility will consist of a query language, a query user interface, a data structure, indexing and finally content searching algorithms. Our interests focus on the first four components of such system. In this thesis, we propose a multimedia query specification language that can be used to fully describe a multimedia segment that needs to be received from a database. We designed an object-oriented database schema that can integrate the logical, temporal and spatial information of a multimedia document. We also proposed an algorithm to index a video by detecting camera breaks and coarse camera operations which will then be used as indices. Moreover, we discussed how to index interactive multimedia documents where undetermined user actions are involved, by proposing a temporal model that has the capability to represent interactive multimedia document scenarios. Indices that represent possible user actions and reactions are then extracted from the model and used to index such documents. Finally, we designed a simple and friendly query user interface on top of the proposed language, in which users can fully describe graphically a multimedia segment.