电子学报 ›› 2017, Vol. 45 ›› Issue (1): 164-172.DOI: 10.3969/j.issn.0372-2112.2017.01.023

• 学术论文 • 上一篇    下一篇

基于几何和纹理特征的表情层级分类方法

胡敏1,2, 江河1,2, 王晓华1,2, 许良凤1, 黄晓音1,2, 程轶红1,2   

  1. 1. 合肥工业大学计算机与信息学院, 安徽合肥 230009;
    2. 情感计算与先进智能机器安徽省重点实验室, 安徽合肥 230009
  • 收稿日期:2015-05-19 修回日期:2016-01-14 出版日期:2017-01-25
    • 通讯作者:
    • 江河
    • 作者简介:
    • 胡敏,女,1967年8月出生,安徽淮北人,教授、硕士生导师,2004年获合肥工业大学计算机应用技术博士学位.主要研究方向:计算机视觉、数字图像处理,等.E-mail:uhnim@163.com

A Hierarchical Classification Method of Expressions Based on Geometric and Texture Features

HU Min1,2, JIANG He1,2, WANG Xiao-hua1,2, XU Liang-feng1, HUANG Xiao-yin1,2, CHENG Yi-hong1,2   

  1. 1. School of Computer and Information, Hefei University of Technology, Hefei, Anhui 230009, China;
    2. Anhui Province Key Laboratory of Affective Computing and Advanced Intelligent Machine, Hefei, Anhui 230009, China
  • Received:2015-05-19 Revised:2016-01-14 Online:2017-01-25 Published:2017-01-25

摘要:

针对表情识别,为提取对个体差异鲁棒性更强的特征,并有效利用特征自身分布特性,本文提出基于几何和纹理特征的表情层级分类方法.首先,构建基于中性脸相似度的几何特征提取方法,自动匹配样本相似中性脸,提取特征点比例系数几何特征;然后,利用充分矢量三角形提取纹理特征;最后,给出表情层级分类框架,在三个层级下分别利用提取特征判定表情类别.所提方法在JAFFE库和CK库上的实验结果表明,本文方法取得了比基于一般几何和纹理特征的识别方法更好的效果,证明了本文方法的有效性.

关键词: 表情识别, 几何和纹理特征, 中性脸相似度, 层级分类

Abstract:

In order to strengthen the robustness of the extracted features for individual differences and use the distribution characteristics of the features more effectively,this paper presents a hierarchical classification method of expression based on geometric and texture features.Firstly,a geometric feature extract method is constructed based on the similarity of neutral expression,which automatically matches with the similar neutral expression images of sample images and extract geometric features based on feature points scale factor.Then,texture features are extracted by using sufficient vector triangle pattern.Finally,the facial expression hierarchical classification framework is achieved by using the above features to judge expression categories in the three layers respectively.Experiment results in JAFFE database and CK database show that the proposed method improves the recognition rate compared with the methods based on the typically geometric and texture features.

Key words: expression recognition, geometric and texture features, similarity of neutral expression, hierarchical classification

中图分类号: