一种变换域纹理图像特征提取和分割方法  

Transform domain method for feature extraction and segmentation of textured image.

作　　者：岳爱菊[1] 汪西莉[1] 

机构地区：[1]陕西师范大学计算机科学学院,西安710062

出　　处：《计算机工程与应用》2007年第31期50-52,57,共4页Computer Engineering and Applications

基　　金：国家自然科学基金(the National Natural Science Foundation of China under Grant No.40671133)

摘　　要：以对偶树复小波变换为基础,提出了一种提取纹理图像变换域统计特征,进而实现图像非监督分割的方法。该方法用Gamma分布和对数正态分布建模对偶树复小波系数的模值,将两种分布的参数综合起来作为像素特征,利用边缘保持的平滑技术(EPNSQ)进行特征平滑,最后使用K-均值聚类方法实现特征分类,得到图像的非监督分割结果。实验结果表明所提取的特征可以有效地表征不同的纹理,基于该特征得到了更为精确的图像分割结果。In this paper,we proposed a method for extracting the texture features in transform domain based on Dual Tree Complex Wavelet Transform（DT-CWT）,and applied it to unsupervised textured image segmentation.The image is first decomposed by one level DT-CWT.We model the modular of complex coefficients with Gamma distribution and lognormal distributrion.Texture features are extracted using Gamma distribution parameter and Lognormal distribution parameter.Then feature vectors are smoothed by the EPNSQ filter.Finally,unsupervised image segmentation results are obtained using K-means clustering based on the features.The experiment results show the effectiveness and precision of the proposed feature extraction and segmentation method.

关 键 词：对偶树复小波变换 Gamma分布参数 对数正态分布参数 特征提取 纹理分割 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]