Author Login Editor-in-Chief Peer Review Editor Work Office Work

Computer Engineering

Previous Articles    

An Image Classification Method Based on High-order Feature Representation

ZHOU Yishu 1,LIANG Peng 2,ZHANG Haitao 1, ZHOU Jianxiong 1,LIU Qi 1   

  1. (1.South Base,China Mobile,Guangzhou 510640,China;2.School of Computer Science,Guangdong Polytechnic Normal University,Guangzhou 510665,China)
  • Received:2015-03-09 Online:2016-04-15 Published:2016-04-15

一种基于高阶特征表示的图像分类方法

周毅书1,梁鹏2,张海涛1 ,周建雄1 ,刘琦1   

  1. (1.中国移动南方基地,广州 510640; 2.广东技术师范学院计算机科学学院,广州 510665)
  • 作者简介:周毅书(1981-),男,高级工程师、硕士,主研方向为大数据应用、云计算;梁鹏(通讯作者),讲师、博士研究生;张海涛、周建雄、刘琦,高级工程师、硕士。
  • 基金资助:
    广东省高等学校科技创新基金资助项目(2013KJCX0117)。

Abstract: The traditional high order feature representation methods are based on feature matching,which can cause combination explosion problem.In this case,this paper proposes a novel high-order feature representation method without feature matching.The image is divided into multiple grids.The scale Invariant Feature Transform(SIFT) features in each grid are combined into one high-order feature representation.The feature of maximum tf-idf value is regarded as the major feature of the grid,while the other as auxiliary features.The angle between,the major feature and the auxiliary features is used as the information of the feature space.High-order feature matching is based on vision matching and geometry matching,and Support Vector Machine(SVM) is used for image classification experiments.Experimental results show that the proposed method can improve the classiffication accuracy by 4%,and effectively avoid the combination explosion problem.

Key words: high-order feature combination, image classification, feature matching, Support Vector Machine(SVM), space structure information

摘要: 传统高阶特征构建方法需对局部特征进行两两匹配,随着阶数的增加,从而导致组合爆炸问题。为此,提出一种高阶特征组合表示方法。将图像划分为多个网格,每个网格内的尺度不变特征转换组合成高阶特征组合。同一个网格内tf-idf值最大的特征作为主特征,其余特征作为子特征,将主特征与子特征之间的夹角作为特征空间信息。通过高阶特征组合的视觉距离和几何距离匹配高阶特征,并结合支持向量机进行图像分类实验,结果表明,该方法分类准确率提高了约4%,可有效避免组合爆炸问题。

关键词: 高阶特征组合, 图像分类, 特征匹配, 支持向量机, 空间结构信息

CLC Number: