Developing AContent-Based Image Retrieval System Using Segmentation and Color Feature.
No Thumbnail Available
Date
2008-03
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Addis Ababa University
Abstract
In this thesis, a Content-Based Image Retrieval (CBIR) system that is based on
segmentation and color feature is presented. CBIR helps to organize and retrieve digital
images using their visual content from large databases. The CBIR system proposed in this
work consists of segmentation, feature extraction and similarity measuring techniques to
store and to retrieve the images.
The segmentation is done using the K-Means clustering algorithm. Unlike existing
systems, the segmentation used is unreliable that the image is divided into many small
regions that are far from representing semantic objects.
Feature extraction is done for each region of the image. A region is represented by its
average color and number of pixels. The average color is calculated for each channel of
the RGB color space. An image is stored in the database as a collection of regions.
The similarity of two images is based on the similarity of their regions. Two different
similarity measures are proposed in this work. One is based on count of similar regions
and the other measure is based on the sum of pixels of the similar regions. Region
similarity is based on the Euclidean distance between their average colors and the number
of pixels of the regions. Two images that have more number of similar regions are said to
be more similar.
The proposed system is compared with some existing systems like Earth Mover’s
Distance (EMD) and SIMPLIcity using three parameters: precision, average rank and
standard deviation of the ranks. The performance of the proposed system is promising
and found to be better than EMD. However, the system didn’t perform well compared to
SIMPLIcity, owing to the fact that SIMPLIcity uses color, shape and texture for the
feature extraction technique whereas the proposed system uses only the color feature.
Description
Keywords
Color Feature