Abstract: Based on the characteristics of good sparsity and capturing effectively the smooth contours in natural images for Contourlet transform.This paper completes texture analysis of the image on the demographic characteristics of Contourlet coefficient and pursues a filter algorithm.According to the statistical characteristics of Contourlet coefficients，Gauss model is used to demonstrate the statistics of those coefficients and extraction feature.The feature vector is weigh ted according to their degrees of the dispersion，and the feature with higher discrimination quality has bigger weight.At the same time，it obtains the final feature vectors by adding the statistic properties of the gray level co-occurrence matrix extracted from the sub band low-frequency.The texture similarities of images are computed by the Euclidean’s distance with weight.Simulation results show that the method perform certain superiority in the retrieval results.

Key words: Contourlet transform, feature extraction, textureretrieva

摘要： 利用 Contourlet 变换良好的稀疏特性及其能准确地捕获图像中边缘信息的特性，分析了纹理图像Contourlet系数的统计特征，提出了一种滤波算法。该算法根据纹理图像Contourlet系数分布的特点，采用高斯模型对其进行精确拟合提取纹理特征，针对各子带数据的离散程度进行加权处理，为分类能力强的特征量赋予较大的权值，并加入在低频子带上提取的灰度—平滑共生矩阵统计量，来形成最终的特征向量，以两幅图像不同特征向量间的加权欧氏距离作为图像的相似度进行检索。仿真实验结果表明，该方法在检索效果上有一定的优越性。

关键词: Contourlet变换, 特征提取, 图像检索