计算机科学与探索 ›› 2017, Vol. 11 ›› Issue (3): 396-405.DOI: 10.3778/j.issn.1673-9418.1603048

• 人工智能与模式识别 • 上一篇    下一篇

利用颜色进行层次模式挖掘的图像分类方法

朱  杰1,2+,超木日力格1,谢博鋆3,于  剑1   

  1. 1. 北京交通大学 计算机与信息技术学院 交通数据分析与挖掘北京市重点实验室,北京 100044
    2. 中央司法警官学院 信息管理系,河北 保定 071000
    3. 河北大学 数学与信息技术学院 机器学习与计算智能重点实验室,河北 保定 071000
  • 出版日期:2017-03-01 发布日期:2017-03-09

Image Hierarchical Pattern Mining and Classification Based on Color

ZHU Jie1,2+, Chaomurilige1, XIE Bojun3, YU Jian1   

  1. 1. Beijing Key Lab of Traffic Data Analysis and Mining, School of Computer and Information Technology, Beijing Jiaotong University, Beijing 100044, China
    2. Department of Information Management, The Central Institute for Correctional Police, Baoding, Hebei 071000, China
    3. Key Laboratory of Machine Learning and Computational Intelligence, College of Mathematics and Information Science, Hebei University, Baoding, Hebei 071000, China
  • Online:2017-03-01 Published:2017-03-09

摘要: 图像的中间层特征挖掘能够发现不同视觉词之间的关系,然后可以利用挖掘得到的模式代替原有的视觉词进行图像表示。目前大部分的中层特征挖掘都是针对所有图像块进行的,而没有考虑到可以在局部进行模式挖掘。在局部进行模式挖掘有利于发现不同对象区域的模式,并且最终提高图像分类的准确率。提出了一种有效的基于颜色的层次模式挖掘方法。该方法把对有判别力的颜色的判断作为划分层次的标准,然后在每一层中对拥有这些颜色的图像块进行挖掘,最后用挖掘到的模式进行图像表示,并用于图像分类。实验结果表明,所提方法能够在Soccer、Flower 17和Flower 102上取得良好的分类效果。

关键词: 有判别力的颜色选择, 中间层特征挖掘, 模式, 分类

Abstract: Mid-level features mining is effective in discovering the relation between different visual words, and then the mined patterns are used to replace the visual words for image representation. All the patches are used to discover the patterns in most of the recent methods, while mining the patterns in local areas are not considered. Mining the patterns in local areas can discover the patterns in different object parts, finally the image representations constructed based on these patterns can help to increase the classification accuracy. This paper proposes an effective color based hierarchical pattern mining method. The discriminative colors are used to divide the images into different levels, and then the patterns are discovered from the patches in each different level, finally the discovered patterns are used for the final image representation. Classification results are presented on Soccer, Flower 17 and Flower 102 datasets, and the experiments demonstrate that the proposed method can obtain satisfactory results on these datasets.

Key words: discriminative color selection, mid-level features mining, pattern, classification