利用特征加权进行基于小波变换的纹理分类  

WAVELET TRANSFORM-BASED TEXTURE CLASSIFICATION USING FEATURE WEIGHTING

作　　者：吴高洪[1] 章毓晋[1] 林行刚[1] 

机构地区：[1]清华大学电子工程系,北京100084

出　　处：《模式识别与人工智能》1999年第3期262-267,共6页Pattern Recognition and Artificial Intelligence

基　　金：国家自然科学基金;863计划;智能技术与系统国家重点实验室资助项目

摘　　要：本文提出了一种利用特征加权进行基于小波变换的纹理分类方法。本方法选用Daubechies正交小波,采用标准的金字塔结构小波变换,将小波变换各个频带输出的l_1范数作为纹理分类的特征,并根据特征本身的离散程度对其进行加权,最后,采用最小距离分类器进行分类。对近千例测试样本的分类实验表明,本文提出的算法与无特征加权算法相比,性能有明显的提高。In this paper, a new approach to wavelet transform-based texture classification using feature weighting is presented. The Daubechies orthogonal wavelets are selected and the standard pyramid-structured wavelet transform is employed. This approach extracts the l1 -norm for each frequency channel of the wavelet transform output as the features for texture classification, and weights these features according to their own degree of dispersion. A minimum distance classifier is used in the classification procedure. The classification experiments for almost 1000 testing samples obtained from 20 classes of textures show that the performance of the presented texture classification algorithm is much better than the algorithm without feature weighting.

关 键 词：纹理分析 纹理分类 小波变换 特征提取 图像分析 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] TN919.8[自动化与计算机技术—计算机科学与技术]