电子学报 ›› 2017, Vol. 45 ›› Issue (8): 1882-1887.DOI: 10.3969/j.issn.0372-2112.2017.08.011

• 学术论文 • 上一篇    下一篇

基于图正则化局部特征编码算法的图像分类方法

杨赛1, 赵春霞2, 胡彬3, 陈峰1   

  1. 1. 南通大学电气工程学院, 南通 226019;
    2. 南京理工大学计算机科学与工程学院, 江苏南京 210094;
    3. 南通大学计算机科学与技术学院, 江苏南通 226019
  • 收稿日期:2016-02-22 修回日期:2016-10-13 出版日期:2017-08-25
    • 通讯作者:
    • 杨赛
    • 作者简介:
    • 赵春霞,女,1964年出生,北京人.1985年、1988年和1998年在哈尔滨工业大学分别获得工学学士、工学硕士和工学博士学位,现为南京理工大学教授,博士生导师.主要研究方向为地面智能机器人与复杂环境理解.
    • 基金资助:
    • 江苏省普通高校自然科学研究面上项目 (No.16KJB520037); 国家自然科学基金 (No.61602150); 江苏省自然科学基金 (No.BK20151273); 南通市科技项目前沿与关键技术 (No.MS22015100); 江苏省博士后科研资助计划项目 (No.1601013B)

An Image Classification Method Using Graphically Regularized Coding Algorithm

YANG Sai1, ZHAO Chun-xia2, HU Bin3, CHEN Feng1   

  1. 1. School of Electrical Engineering, Natong University, Nantong, Jiangsu 226019, China;
    2. School of Computer Science and Engineering, Nanjing University of Science and Technology, Nanjing, Jiangsu 210094, China;
    3. School of Computer Science and Technology, Nantong University, Nantong, Jiangsu 226019, China
  • Received:2016-02-22 Revised:2016-10-13 Online:2017-08-25 Published:2017-08-25
    • Supported by:
    • General Program of Natural Science Research in Colleges and Universities of Jiangsu Province (No.16KJB520037); National Natural Science Foundation of China (No.61602150); Natural Science Foundation of Jiangsu Province,  China (No.BK20151273); Frotier and Key Technique of Nantong Science and Technology Program (No.MS22015100); Funded by Postgraduate Science Research Program of Jiangsu Province (No.1601013B)

摘要: 为了解决经典局部特征编码方法会产生相似局部特征之间编码系数不一致的问题,本文提出一种图正则化局部特征编码算法.该算法在对初始编码矢量所定义的能量化函数中引入正则化项,保证空间上相邻外观上相似的局部特征之间的编码矢量尽可能一致.MSRcv2、Caltech101、Scene 15以及Indoor 67四个公开数据集上的实验结果表明本文方法能够提高硬分配、软分配、稀疏编码、局部约束线性编码以及局部软分配五种经典编码方法的性能,并且基于本文编码算法的图像分类方法在上述四个公开数据集上的平均分类正确率分别达到了91.13%、76.02%、83.76%、44.78%.

关键词: 词袋模型, 编码算法, 图模型, 图像分类

Abstract: In order to solve the problem that current coding schemes lost consistence between similar local features,this paper proposes a new graphically regularized coding algorithm.This algorithm used any current coding scheme to get the initial coding coefficients,and utilized a regularized term to preserve locality constrains both in the feature space and the spatial domain of the image.Experimental results on popular benchmark datasets show that our method improves the performances of the current coding algorithms,and the average classification accuracies of our proposed method in MSRcv2,Caltech101,Scene15,Indoor 67 and UIUC-sport has reached 91.13%,76.02%,83.76%,44.78% and 89.05% respectively.

Key words: bag-of-feature, coding algorithm, graphical model, image classification

中图分类号: