direct pic.gif

Project Info

Title Decentralized Image Retrieval for Education (DIRECT)
Goal Build a image based retrieval system for the image data bases of NASA libraries
Funding National Science Foundation (NSF)


To build a decentralized retrieval system which can retrieve the images from different databases. The retrieval is based on segmenting the query image into objects first and then compare its characteristics with those of objects from other images in the database.


Project DIRECT is a new and different content based image retrieval system. The data base of images is decentralized. The retrieval is based on segmenting each image in the data base into meaningful objects then extracting and storing certain characteristics of the objects like shape, color, and texture. To find similiar images, a query image is segmented into objects and those characteristics are compared to the characteristic objects in the images to be searched. A matching criterion is formed. The images are ranked and the most similar images are displayed.

Segmentation. The given image is converted into YCbCr image. The image is then classified into five classes using k-means. In each class, regions are identified using connected component labeling. The area of these regions are calculated and the biggest region in each class is selected as the dominant segment.

Shape: The perimeter of the dominant segment in each class is found and the Fourier shape descriptor coefficients are calulated and stored. Currently we are working to continue improving the segmentation part of the project.


Topic revision: r2 - 30 May 2014, AndreaVaccari
©2018 University of Virginia. Privacy Statement
Virginia Image and Video Analysis - School of Engineering and Applied Science - University of Virginia
P.O. Box 400743 - Charlottesville, VA - 22904 - E-Mail viva.uva@gmailREMOVEME.com