Color Image Classification.

Date of Submission

December 2007

Date of Award

Winter 12-12-2008

Institute Name (Publisher)

Indian Statistical Institute

Document Type

Master's Dissertation

Degree Name

Master of Technology

Subject Name

Computer Science

Department

Machine Intelligence Unit (MIU-Kolkata)

Supervisor

Murthy, C. A. (MIU-Kolkata; ISI)

Abstract (Summary of the Work)

In this dissertation, we were given a data set, and the problem is to classify each image into one of the classes. The data set is named test1.tar, and located at the web site wang.ist.psu.edu/docs/related .The data set contains ten classes of images. Each class contains hundred images. All images are color images.The names of ten classes are:1. Flower.2. Dinosaur.3. Elephant.4. Horse.5. Mountain.6. Tribal.7. Building.8. Bus.9. Food items.10. Beach.Our problem is to distinguish between the first eight classes of images in the above list, that is, if we are given one image among these eight classes then the machine would automatically classify it properly into one of the eight classes.1.1 Characteristics of dataIn this section, we describe the characteristics of the object classes in our data set. One main characteristic is, except the ‘Dinosaur’ class images, all other class images are ‘natural’ images taken from the real world. The ‘Dinosaur’ class images are ‘artificial’ images.Experts artificially make the images of the ‘Dinosaur’ class. All the images of this class have white background and the color of the bottom portion of the dinosaur in the image is close to the white color. However, the dinosaurs have different colors. In almost all of the ‘Dinosaur’ images, less number of colors dominates an object (dinosaur) in the image. There is only one dinosaur in an image.The color of a flower in the ‘Flower’ class images is uniform. Usually green leaves cover the background in this class images. In most of the images, background is blurred. There are at most two flowers in most of the images. The color of all the flowers is same in an image.In the ‘Bus’ class images, most of them have only one bus in an image, but in some images there are two; one bus is partially occluded by other, that is, one bus is completely visible and the other bus is partially occluded by this bus. The background is not uniform in all images. In some images it may be covered by trees and sky and in some others, a building may be there. Important property in ‘Bus’ class images is that they contain regular shapes, buses are painted with different colors but in regular shapes, the windows of the bus are also in regular shapes (e.g. rectangular, circular, etc.).We considered less number of images for the class ‘Building’. We ignored the images that do not enough features in it. In the ‘Building’ class images, some of the images contain human beings in front of the buildings, but they are small when compared to the size of the building. In some images, the background is covered with sky, whereas in some others, the building covers the whole image. ‘Building’ class images contain regular shapes (e.g. rectangular, cylindrical).In the ‘Elephant’ class images; there may be more than one elephant in the image. As we, all know elephants are of grey color. An elephant may be in between trees, it may be at a pond, or may be standing on clay. In most of the images, the tusk of the elephant is visible. Trunk is also visible in almost all the images.

Comments

ProQuest Collection ID: http://gateway.proquest.com/openurl?url_ver=Z39.88-2004&rft_val_fmt=info:ofi/fmt:kev:mtx:dissertation&res_dat=xri:pqm&rft_dat=xri:pqdiss:28843718

Control Number

ISI-DISS-2007-210

Creative Commons License

Creative Commons Attribution 4.0 International License
This work is licensed under a Creative Commons Attribution 4.0 International License.

DOI

http://dspace.isical.ac.in:8080/jspui/handle/10263/6373

This document is currently not available here.

Share

COinS