一种局部频域分析的多尺度纹理特征提取方法  

A Multi-scale Texture Classification Method of Local Frequency Domain Analysis

作　　者：徐敏[1] 张哲铭[1] 周中伟[2] 

机构地区：[1]厦门理工学院电气工程与自动化学院,福建厦门361024 [2]江西理工大学电气工程与自动化学院,江西赣州341000

出　　处：《厦门理工学院学报》2014年第5期35-40,共6页Journal of Xiamen University of Technology

基　　金：福建省科技厅科技计划重点项目(2014H0047);福建省教育厅科技项目(JA12247);厦门市科技计划项目(3502Z20143034)

摘　　要：在局部二进制模型和多分辨率分析的基础上,提出一种基于局部频域分析的多尺度纹理特征提取方法.对纹理图像采用局部圆形邻域作为局部模板进行局部频率变换,对各频段图像实现多分辨率分析,提取多尺度下的特征向量,结合SVM算法进行分类实验.结果证明,该方法在继承了LBP对光照因素具有的稳定性的同时,具有旋转不变性特征;利用NSCT在多尺度下进行的纹理特征提取也满足尺度不变性的需要,与其他方法相比提高了分类准确率.Based on the analysis of local binary pattern and multi-resolution, this paper presented a multi-scale texture classification method based on local frequency operator. The local frequency conversion was performed in the original image, and each sub-band image had multi-resolution analysis and multi-scale extraction of feature vectors. Classification experiments combining SVM algorithm was then realized. Experimental results show that this method inherits stability of LBP to light factors, while maintaining features to be invariant to rotation. The texture feature extraction meets the needs of scale invariance by using NSCT, and enhances the classification accuracy compared with other methods.

关 键 词：纹理分类 局部频域 局部二进制模式 不变性 纹理特征提取 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]