Date of Award

Winter 2002

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Computer Science

Committee Director

Hussein Abdel-Wahab

Committee Member

Kurt Maly

Committee Member

Shunichi Toida

Committee Member

Christian Wild

Committee Member

Lee Belfore

Abstract

Multimedia applications are rapidly spread at an ever-increasing rate introducing a number of challenging problems at the hands of the research community, The most significant and influential problem, among them, is the effective access to stored data. In spite of the popularity of keyword-based search technique in alphanumeric databases, it is inadequate for use with multimedia data due to their unstructured nature. On the other hand, a number of content-based access techniques have been developed in the context of image indexing and retrieval; meanwhile video retrieval systems start to gain wide attention, This work proposes a number of techniques constituting a fully content-based system for retrieving video data. These techniques are primarily targeting the efficiency, reliability, scalability, extensibility, and effectiveness requirements of such applications. First, an abstract representation of the video stream, known as the DC sequence, is extracted. Second, to deal with the problem of video segmentation, an efficient neural network model is introduced. The novel use of the neural network improves the reliability while the efficiency is achieved through the instantaneous use of the recall phase to identify shot boundaries. Third, the problem of key frames extraction is addressed using two efficient algorithms that adapt their selection decisions based on the amount of activity found in each video shot enabling the selection of a near optimal expressive set of key frames. Fourth, the developed system employs an indexing scheme that supports two low-level features, color and texture, to represent video data, Finally, we propose, in the retrieval stage, a novel model for performing video data matching task that integrates a number of human-based similarity factors. All our software implementations are in Java, which enables it to be used across heterogeneous platforms. The retrieval system performance has been evaluated yielding a very good retrieval rate and accuracy, which demonstrate the effectiveness of the developed system.

DOI

10.25777/mg4g-7d44

ISBN

9780493883120

Share

COinS