Texture Image Retrieval Based on Contourlet Coefficient Modeling with Generalized Gaussian Distribution

Abstract

This paper presents a texture image retrieval scheme based on contourlet transform. In this scheme, the generalized Gaussian distribution (GGD) parameters are used to represent the detail subband features obtained by contourlet transform. To obtain these parameters, an improved maximum likelihood (ML) parameter estimation method is proposed, in which a new initial estimation value is exploited and a modified iterative algorithm is used. Compared with existing features used for the texture image retrieval, the use of the GGD parameters to represent the contourlet detail subbands provides richer information to improve the retrieval accuracy. The proposed retrieval scheme is demonstrated on the VisTex database of 640 texture images. Experimental results show that, compared with the current ML estimation and texture retrieval method, the proposed scheme can give more accurate estimates of the GGD parameters, and it improves more effectively the average retrieval rate from 76.05% to 78.09% with comparable computational complexity.