关键词: Contourlet变换, 纹理图像检索, 广义高斯分布, 参数估计

Abstract: Contourlet transform has been widely applied to texture analysis due to its providing rich directional information.In this paper，an improved contourlet transform which has sharply localized basis images in the frequency domain is applied for texture image retrieval.Its directional subband coefficients are modeled as Generalized Gaussian Distribution（GGD），and the distribution parameters are used as the features of texture image retrieval，which are estimated by using an improved maximum likelihood method.The texture images are divided into the structured and random texture images based on Shannon entropy measurement，and a novel hybrid retrieval scheme based on energy feature and GGD parameters is proposed.Experimental results based on the VisTex database with 640 texture images show that，compared with the method based on the discrete wavelet transform，the average retrieval rate is improved from 76.57% to 83.55%.

Key words: Contourlet transform, texture image retrieval, generalized Gaussian distribution, parameter estimation