计算机工程 ›› 2012, Vol. 38 ›› Issue (19): 175-178.doi: 10.3969/j.issn.1000-3428.2012.19.045

• 人工智能及识别技术 • 上一篇    下一篇

基于神经网络和主元分析的人脸识别算法

何正风  1,孙亚民  2   

  1. (1. 佛山科学技术学院基础教育系,广东 佛山 528000;2. 南京理工大学计算机科学与技术学院,南京 210094)
  • 收稿日期:2011-11-30 出版日期:2012-10-05 发布日期:2012-09-29
  • 作者简介:何正风(1957-),男,讲师,主研方向:数学建模仿真,智能计算;孙亚民,教授、博士生导师
  • 基金项目:
    广东省自然科学基金资助项目(S2011020002719, 10152800001000016)

Face Recognition Algorithm Based on Neural Network and Principal Component Analysis

HE Zheng-feng 1, SUN Ya-min 2   

  1. (1. Department of Basic Education, Foshan University, Foshan 528000, China; 2. School of Computer Science and Technololgy, Nanjing University of Science and Technology, Nanjing 210094, China)
  • Received:2011-11-30 Online:2012-10-05 Published:2012-09-29

摘要: 针对高维、小样本的分类问题,提出2个重要的准则,用于估计RBF单元的初始宽度。采用主成分分析方法把训练样本集投影到特征脸空间,以减少维数,用Fisher线性判别式产生一组最具判别性的特征,使不同类间的训练数据尽可能地分开,而同一类的样本尽可能地靠近。实验结果证明,该算法在分类的错误率及学习的效率上都表现出较好的性能。

关键词: 人脸检测, 特征提取, 人脸识别, 聚类算法, 神经网络, 主元分析

Abstract: According to the high dimension, small sample classification problem, this paper puts forward two important criterions to estimate the initial width of RBF unit. Principal Component Analysis(PCA) method used the training sample set is projected onto the eigenface space, in order to reduce the dimensionality, using Fisher linear discriminant to generate a group of the most discriminant features, different classes of the training data can be separated as much as possible, and the same samples are as close as possible. The results prove that this algorithm both in the classification error rate or in the learning efficiency can show excellent performance.

Key words: face detection, feature extraction, face recognition, clustering algorithm, neural network, principal component analysis

中图分类号: