Content-Based Image Retrieval (Project for ECE 532)

Abstract

Content-Based Image Retrieval (CBIR) has received considerable attention in recent years, given the exponential increase in the number of images and digitized documents being used. The main focus of this work is on the evaluation of different features for efficient CBIR.

Various features such as color, chromaticity moments, texture and segment based features such as location, compactness, moments etc. have been investigated. Retrieval using various features has been studied and their usefulness for the task of image retrieval ascertained. The effect of variation in various feature specific parameters is also investigated.

The system has been tested on a database of more than 2000 images from the Corel and Washington Ground Truth Image Databases. 

The entire project has been coded in C and the crawler (for automatically crawling the image databases and calling appropriate binaries) has been coded in Python. All code has been written by me (except that for segmenting images).

Some Sample Retrievals