作者投稿和查稿 主编审稿 专家审稿 编委审稿 远程编辑

计算机工程

• 图形图像处理 • 上一篇    下一篇

基于Gist 和PHOG 特征的场景分类

刘 静,郭 建,贺遵亮   

  1. (湘潭大学材料与光电物理学院,湖南湘潭411105)
  • 收稿日期:2014-05-14 出版日期:2015-04-15 发布日期:2015-04-15
  • 作者简介:刘 静(1987 - ),男,硕士研究生,主研方向:图像处理,模式识别;郭 建,教授;贺遵亮,硕士研究生。

Scene Classification Based on Gist and PHOG Feature

LIU Jing,GUO Jian,HE Zunliang   

  1. (Faculty of Materials,Optoelectronics and Physics,Xiangtan University,Xiangtan 411105,China)
  • Received:2014-05-14 Online:2015-04-15 Published:2015-04-15

摘要: 局部Gist 方法提取的特征维数过高、计算复杂,单一的Gist 特征不能很好地描述全局场景。为此,提出一种将改进的局部Gist 特征与梯度方向直方图特征进行组合的场景描述方法。采用支持向量机作为分类器,在WS场景库中考察单一特征和组合特征的分类精度,在OT 场景库下研究不同数量训练样本对于分类精度的影响。实验结果表明,与全局Gist、局部Gist 等方法相比,该方法能降低计算的复杂度,且提高分类正确率。

关键词: 局部Gist 特征, 梯度方向直方图, 特征组合, 场景描述, 支持向量机, 场景分类

Abstract: In view of complex computation caused by extracting high dimension characteristics with local Gist method, as well as the problem that the sole Gist characteristic can not describe global scenes well,a kind of improved method to describe the scenes is proposed,which combines local Gist characteristics with Histograms of Oriented Gradient(HOG) characteristics. Classification accuracy of the sole characteristics and the combination of characteristics are inspected in the WS scene database using Support Vector Machine(SVM) as the classifier. On this basis,classified precision influenced by different quantity training samples is also studied in the OT scenes database. Experimental results show that this method reduces the computational complexity,and improves the classified accuracy compared with the global Gist,local Gist methods,etc.

Key words: local Gist feature, Histograms of Oriented Gradient(HOG), feature combination, scene description, Support Vector Machine(SVM), scene classification

中图分类号: