Spatio-temporal content characterization and retrieval in multimedia databases
Abstract
The increasing use of multimedia data makes it crucial to use intelligent search mechanisms for retrieving multimedia data by content. Traditional text-based methods clearly do not suffice to describe the rich content present in digital video. We present a video database indexing and search engine, PICTURESQUE (Pictorial Information and Content Transformation Unified Retrieval Engine for Spatiotemporal QUE ries) that incorporates object motion and temporal relationship information in a video database. Spatial composition of the salient objects and their movements play a crucial role in characterization of the “content” of video. PICTURESQUE offers a unified spatio-temporal model and a retrieval scheme for fast and effective access to the underlying database. We propose three different motion representation schemes, trail-based, trajectory coordinates, and interval-based models and validate their effectiveness by comparative performance measures. Our query mechanism is based on an intuitive QBE-like method that involves sketching of the desired motion by the users. Our use of efficient numerical techniques to describe both the query and data composition enables us to take advantage of analytical methods for query evaluation. In addition to improved performance, these methods provide benefits such as flexibility and temporal scale invariance. PICTURESQUE covers a complete solution from pre-indexing of video to presentation of results via a graphical user interface and its modular implementation provides concision and extensibility. Development of such technology will enable true multimedia search engines that will accomplish what current textual search engines do today.
Degree
Ph.D.
Advisors
Ghafoor, Purdue University.
Subject Area
Electrical engineering|Computer science|Information science
Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server.