MCM-CBIR: Multi Clustering Method for Content Based Image Retrieval

Abstract

Image retrieval systems are designed to provide the ability of searching and retrieving images in huge image databases. A content based image retrieval system (CBIR) is used to offer such tasks based on the content of the image. In this paper we propose a new method of CBIR system based on a learning technique. Our method uses k-means clustering to reduce data and to improve the system performance. The specificity of our approach is the use of each feature vector separately in the clustering process in order to obtain different clustering on the same database, differently to other approaches that combine features vectors to cluster the database. For this reason we call it multi-clustering approach. The advantage of this approach consists in keeping the performance of the features and getting several views of the database due to the separation of features. The experimental results show the efficiency of our approach.