Logo image
Utilising fuzzy rough set based on mutual information decreasing method for feature reduction in an Image Retrieval System
Book chapter

Utilising fuzzy rough set based on mutual information decreasing method for feature reduction in an Image Retrieval System

M. Shahabi Lotfabadi, M.F. Shiratuddin and K.W. Wong
Innovations and Advances in Computing, Informatics, Systems Sciences, Networking and Engineering, Vol.313, pp.177-184
Springer International Publishing
2014
url
Link to Published Version *Subscription may be requiredView

Abstract

Content-Based Image Retrieval (CBIR) system has become a focus of research in the area of image processing and machine vision. General CBIR system automatically index and retrieve images with visual features such as colour, texture and shape. However, current research found that there is a significant gap between visual features and semantic features used by humans to describe images. In order to bridge the semantic gap, some researchers have proposed methods for managing and decreasing image features, and extract useful features from a feature vector. This paper presents an image retrieval system utilising fuzzy rough set based on mutual information decreasing method and the Support Vector Machine (SVM) classifier. The system has training and testing phases. In order to reduce the semantic gap, the propose retrieval system used relevance feedback to improve the retrieval performance. This paper also compared the proposed method with other traditional retrieval systems that use PCA, kernel PCA, Isomap and MVU for their feature reduction method. Experiments are carried out using a standard Corel dataset to test the accuracy and robustness of the proposed system. The experiment results show the propose method can retrieve images more efficiently than the traditional methods. The use of fuzzy rough set based on mutual information decreasing method, SVM and relevance feedback ensures that the propose image retrieval system produces results which are highly relevant to the content of an image query.

Details

Metrics

59 Record Views
Logo image