计算机工程 ›› 2012, Vol. 38 ›› Issue (10): 151-153.doi: 10.3969/j.issn.1000-3428.2012.10.046

• 人工智能及识别技术 • 上一篇    下一篇

改进的二维典型相关分析及其人脸识别应用

刘艳艳 1,曹慧荣 2,王建国 3,赵宜宾 1   

  1. (1. 防灾科技学院基础部,河北 三河 065201;2. 廊坊师范学院数学与信息科学学院,河北 廊坊 065000; 3. 北京信息职业技术学院媒体制作中心,北京 100015)
  • 收稿日期:2011-07-20 出版日期:2012-05-20 发布日期:2012-05-20
  • 作者简介:刘艳艳(1980-),女,博士,主研方向:模式识别,数据分析,图像处理;曹慧荣,硕士;王建国,工程师;赵宜宾,副教授
  • 基金项目:
    中国地震局教师科研基金资助项目(20110116);河北省自然科学基金资助项目(A2011408006)

Improved Two-dimensional Canonical Correlation Analysis and Its Application in Face Recognition

LIU Yan-yan 1, CAO Hui-rong 2, WANG Jian-guo 3, ZHAO Yi-bin 1   

  1. (1. Department of Basic Courses, Institute of Disaster Prevention Science and Technology, Sanhe 065201, China; 2. College of Mathematics and Information Science, Langfang Normal College, Langfang 065000, China; 3. Multimedia Making Center, Beijing Information Technology College, Beijing 100015, China)
  • Received:2011-07-20 Online:2012-05-20 Published:2012-05-20

摘要: 针对二维典型相关分析(2DCCA)中类标矩阵维数较大及算法耗时过多的问题,提出一种改进的2DCCA特征提取方法。利用图像的频谱性质定义低维的类标矩阵,从有利于模式分类的角度构造出新的准则函数,采用二维主成分分析对所得特征进一步降维,得到更具分类判别能力的低维特征。在ORL和组合人脸数据库上的实验结果表明,该特征具有较好的分类能力。

关键词: 二维典型相关分析, 频谱特征, 类标矩阵, 准则函数, 特征提取, 人脸识别

Abstract: An Enhanced Two-dimensional Canonical Correlation Analysis(E-2DCCA) method is presented to solve the problem that 2DCCA requires much storage space and runtime. By making use of the spectrum representation of images, a new class-membership matrix is constructed. A modified correlation criterion function is proposed from the angel of favoring classification. Two-dimensional Principal Component Analysis(2DPCA) method is used for further dimensional reduction. Experimental results on ORL and combined face databases show that the features have powerful ability of recognition.

Key words: Two-dimensional Canonical Correlation Analysis(2DCCA), spectrum feature, class-membership matrix, criterion function, feature extraction, face recognition

中图分类号: