Contourlet-S and Generalized Gaussian Model Texture Image Retrieval System

Abstract

To improve the retrieval rate of contourlet transform texture image retrieval system, a new texture image retrieval system was proposed. In the system, contourlet-S, which was a combination of non-subsampled Laplacian Pyramid and critical subsampled directional filter banks, was used to extract directional information of different scales. Generalized Gaussian Density (GGD) model parameters were cascaded to form feature vectors and Kullback-Leibler distance (KLD) function was used for similarity measure. Experimental results on 640 texture images from Vistex texture image database indicate that contourlet-S transform based image retrieval system is superior to that of the original contourlet transform and non-subsampled contourlet transform under the same system structure with almost same length of feature vectors, retrieval time and memory needed. Furthermore, decomposition parameters including the number of scale and directional subband on each scale selected in every contourlet transform can make effects on retrieval rates.