基于局部特征描述和纹理基元学习的纹理分类  

Texture Classification Based on Local Feature Description and Texton Learning

作　　者：王军敏 樊养余[1] 李祖贺[1] WANG Junmin;FAN Yangyu;LI Zuhe(School of Electronics and Information,Northwestern Polytechnical University,Xi＇an 710129;School of Information Engineering,Pingdingshan University,Pingdingshan 467000)

机构地区：[1]西北工业大学电子信息学院,西安710129 [2]平顶山学院信息工程学院,平顶山467000

出　　处：《计算机与数字工程》2018年第9期1861-1865,共5页Computer & Digital Engineering

基　　金：国家自然科学基金项目(编号:61702462);国家质量监督检验检疫总局科技计划项目(编号:2016QK169)资助

摘　　要：为了提高纹理分类算法的性能,提出一种基于局部特征描述和纹理基元学习的纹理分类算法。首先利用多个局部特征量描述每个像素的局部特征,然后利用K均值聚类算法从局部特征中学习纹理基元字典,再利用纹理基元字典对纹理图像进行编码,并计算编码图像的纹理基元直方图作为纹理图像的特征向量,最后采用最近子空间分类器进行分类。在CURe T和KTH-TIPS纹理库上的实验结果表明,论文算法不但能获得最高的分类精度,并且具有较高的实时性。In order to enhance the performance of texture classification approaches,a texture classification method based on local feature description and texton learning is proposed. First,multiple local descriptors are used to describe the local feature of each pixel of the texture image. Second,K-means clustering algorithm is used to learn the textons from each class of textures,and the texton dictionary is constructed by concatenating the learned textons. Third,each texture image is encoded by the texton dictionary,and the texton histogram is used as the feature description of the texture image. Finally,the nearest subspace classifier（NSC）is used to implement the texture classification. The experimental results on CURe T and KTH-TIPS databases show that the proposed method can achieve higher classification performance with lower time cost.

关 键 词：纹理分类 特征提取 纹理基元学习 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]