Document Type

Thesis

Degree

Master of Science

Major

Computer Science

Date of Defense

4-19-2006

Graduate Advisor

Sanjiv K. Bhatia, Ph.D.

Committee

Martin Pelikan

Cezary Janikow

Abstract

This thesis advances RISE (Robust Image Search Engine), an image database application designed to build and search an image repository. rise is built on the foundation of a CBIR (Content Based Image Retrieval) system. The basic goal of this system is to compute content similarity of images based on their color signatures. The color signature of an image is computed by systematically dividing the image into a number of small blocks and computing the average color of each block using ideas from DCT (Discrete Cosine Transform) that forms the basis for JPEG (Joint Photographic Experts Group) compression format. The average color extracted from each block is used to construct a tree structure and finally, the tree structure is compared with similar structures already stored in the database. During the query process, an image is given to the system as a query image and the system returns a set of images that have similar content or color distribution as the given image. The query image is processed to create its signature which is then matched against similar signature of images already stored in the database. The content similarity is measured by computing normalized Euclidean distance between the query image and the images already stored in the database. RISE has a GUI (Graphic User Interface) front end and a Java servlet in the back end that searches the images stored in the database and returns the results to the web browser. RISE enhances the performance of image operations of the system using JAI (Java Advance Imaging) tools.

Share

COinS