Content Based Image Retrieval Using Self Organizing Map

Abstract

From the recent literature it is observed that information storage and retrieval through the Internet has made impressive progress. Practical searching for information still confronts us with retrieval systems that are present. A Content Based Image Retrieval (CBIR) system provides an efficient way of retrieving related images from image collections. In this paper we present a new feature extraction techniques and clustering of the features to achieve better performance in image retrieval system. The proposed method uses an approach which combines edge information and median filtering technique to extract the features from the image. Self Organizing Map (SOM) technique is used for clustering the extracted image features. The median filtering technique is applied to the original image to get a smooth image. The edge information can be extracted from the image using Bi-directional Empirical Mode Decomposition (BEMD) technique. Then replace only the values of edge position of smooth image with the detected edge image values by BEMD and extracted only 64 bins gray features. These extracted features are supplied as input to the SOM neural network for clustering where features are clustered into nine different groups. Finally query image features are feed to the neural network to identify the cluster to which the query image belongs. The surrounded clustered features are compared with the query image features and display the similar resultant images. The experiment is carried out on a ground truth database which has 1000 images of different categories. The experimental results have been compared with the conventional Median filter histogram technique. Here performance of the retrieval system is good because of combination of median, edge and SOM techniques. It gives an average precision 2.37 % and recall 2.82 % improvement compared with an existing system.