Content Based Image Retrieval by Preprocessing Image Database

Kommineni, J and C, Krishna Mohan (2011) Content Based Image Retrieval by Preprocessing Image Database. Masters thesis, Indian Institute of Technology, Hyderabad.

Preview

Text
CS09G002.pdf - Submitted Version
Download (5MB)

Abstract

Increase in communication bandwidth, information content and the size of the multimedia databases have given rise to the concept of Content Based Image Retrieval (CBIR). Content based image retrieval is a technique that enables a user to extract similar images based on a query, from a database containing a large amount of images. A basic issue in designing a content based image retrieval system is to select the image features that best represent image content in a database. Current research in this area focuses on improving image retrieval accuracy. In this work, we have presented an ecient system for content based image retrieval. The system exploits the multiple features such as color, edge density, boolean edge density and histogram information features. The existing methods are concentrating on the relevance feedback techniques to improve the count of similar images related to a query from the raw image database. In this thesis, we propose a dierent strategy called preprocessing image database using k means clustering and genetic algorithm so that it will further helps to improve image retrieval accuracy. This can be achieved by taking multiple feature set, clustering algorithm and tness function for the genetic algorithms. Preprocessing image database is to cluster the similar images as homogeneous as possible and separate the dissimilar images as heterogeneous as possible. The main aim of this work is to nd the images that are most similar to the query image and new method is proposed for preprocessing image database via genetic algorithm for improved content based image retrieval system. The accuracy of our approach is presented by using performance metrics called confusion matrix, precison graph and F-measures. The clustering purity in more than half of the clusters has been above 90 percent purity.

[error in script]

IITH Creators: