基于NSCT子带融合特征的纹理材质分类  

TEXTURE MATERIAL CLASSIFICATION BASED ON NSCT SUB-BAND FUSION FEATURES

作　　者：陈旭[1] 徐超义 Chen Xu;Xu Chaoyi(School of Automation,Nanjing University of Information Science&Technology,Nanjing 210044,Jiangsu,China)

机构地区：[1]南京信息工程大学自动化学院,江苏南京210044

出　　处：《计算机应用与软件》2023年第2期217-222,264,共7页Computer Applications and Software

基　　金：江苏省自然科学基金项目(BK20170955)。

摘　　要：由于图像纹理特征具有复杂且没有规律的特点,单一的纹理特征提取算法不能充分地描述图像纹理。因此,提出基于NSCT的高低频子带的CS-LBP和Tamura融合特征的纹理分类方法。此外,对CS-LBP算法进行了优化,得到DCS-LEBP,在提取了图像局部边缘特征信息的基础上又加入了低频信息,提高了算法的抗噪性和特征提取的全面性。融合的纹理特征通过支持向量机(SVM)进行分类识别,实验结果表明,该方法能够提取多尺度、多方向的纹理特征,提高了分类精度,并且优化的DCS-LEBP算法比原算法性能更加优秀。Due to the complex and irregular image texture features, a single texture feature extraction algorithm cannot fully describe the texture. Therefore, a texture classification method based on the CS-LBP algorithm with high and low frequency sub-band of NSCT and Tamura texture feature is proposed. In addition, the CS-LBP algorithm was optimized to obtain DCS-LEBP. While considering the local edge information of high frequency, low frequency information was added to improve the noise resistance and feature extraction comprehensiveness of the algorithm. The fused texture features were classified and identified by support vector machine(SVM). The experimental results show that this method can extract multi-scale and multi-direction texture features, improve the classification accuracy, and the optimized DCS-LEBP algorithm has better performance than the original algorithm.

关 键 词：纹理特征 NSCT CS-LBP Tamura 支持向量机 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]