Semantic content-based access to hypervideo databases
Abstract
Technological advances have spurred the use of digital video and generated vast amount of video repositories. The unique characteristics of digital video pose great research challenges to video data management for efficient and effective user access. Recent years have seen increasing research activities in the field of video databases. However, current efforts, especially those on video data modelling, have some of the following shortcomings: (1) they often focus on the visual content and therefore lack modeling of semantic content and spatio-temporal characteristics; (2) the semantic association among video data is not captured; (3) they often depend on relational database schema with fixed sets of attributes and lack generality and flexibility; (4) video objects usually are not fully represented. Due to the weaknesses of video data models, the corresponding query languages are also limited. These factors make user access to video database less than optimal. The goal of this work to provide a practical solution for efficient and effective semantic content-based user access to video databases. A novel video data model called Logical Hypervideo Data Model (LHVDM) is proposed. The model is based on a video abstraction hierarchy and semantic content descriptions. The multi-level data abstractions provide data independence, multi-user view sharing, and data reuse. We define the concept and representation of hot video objects, which is an integrated part of the LHVDM model. Semantic associations among different logical video entities such as hot video objects are captured by video hyperlinks. Based on LHVDM, we further present a video query language that allows users to query and retrieve video based on content descriptions with spatial and temporal constraints. The LHVDM model also supports a user definable and adaptive way of browsing the video database. Video data browsing is done not only on the visual information through progressive multi-level video wrapper but also on the semantic content through user-adaptive video hyperlinks. Finally, a web-based distributed video database prototype is built to demonstrate the soundness of the proposed approach. Several implementation techniques such as lazy delivery, distributed sub-query caching, and user profiling are presented.
Degree
Ph.D.
Advisors
Elmagarmid, Purdue University.
Subject Area
Computer science
Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server.