A METHOD OF IMAGE RETRIEVAL USING BALANCE TREE

Abstract

The paper approaches a method of image retrieval using balance tree to store meta-data of image features by index to describe the shape of interest objects of image. In order to extract interest objects, we propose the image segmentation method on the base of low-level visual features including color and texture of image. The features are extracted at each block of image by Wavelet transformation and CIE L*a*b* color space. From there, we cluster all pixels to give connected regions; at the same time, we remove regions that have area less than threshold. Based on extracted features, we find out the contrast of image in order to extract background and foreground. On the base of segmented image, the feature indexes are created in order to match similar images. Then, we create the balance tree to store the feature indexes and search the set of similar images based on this structure. To illustrate the proposed method, the paper builds application and assesses experimental results on image databases including Corel and Image Object (MSRC). The results of our method are compared with others and show the effective of our proposed method.

Keywords: Segmentation method, interest object, feature index.