Contourlet变换结合了不可分离的方向滤波组,具备小波所不能表达的多方向特性,能有效捕获自然图像的边缘轮廓信息。本文分析了图像Contourlet系数的统计特征,并利用广义高斯函数对各子带系数层进行建模。将此模型应用于基于V isTex的自建纹理图像库,采用矩匹配估计法,提取模型参数集,运用K-L距离计算图像间的相似度。对800幅纹理图像进行检索,本文方法比传统小波方法的平均检索查准率高出约2%到10%不等。实验结果表明,该方法改进了导向纹理的描述。

Combining non-separable and directional filters banks,contourlet transform can effectively capture more edges and contours in natural images than wavelets do due to its capability of representing directional information.This paper casts light on the statistical features of contourlet coefficients,according to which we set up a model using Generalized Gaussian Density Function.To test this model,we applied it in texture images selected from VisTex database.After the extraction of model parameters using moment matching method,Kullback-Leibler(K-L) Distance is used to measure the similarity between images.Experiments on 800 texture images demonstrate that the average retrieval rate using our method is about 2% to 10% higher than that of wavelet method.The method proposed improves the extraction of directional textures.