摘要
标准KCCA方法需要存储和计算核矩阵,而核矩阵的大小是训练样本数的平方,随着样本数的增加,计算量逐渐增大、特征提取缓慢。为了提高特征提取的效率,提出了一种基于特征向量集的KCCA特征提取方法。采用特征选择方法,选择一个训练样本子集并将其映射到再生核希尔伯特空间(RKHS)。用KCCA进行特征提取,将计算复杂度由3降到2(<<),并将改进后的KCCA与SVDD的优势相结合应用于人脸识别中。实验结果表明,相对传统的KCCA方法,所提出的方法在不影响识别率的前提下,显著提高了人脸识别速度,减小了系统的存储量。
According to the standard KCCA-based extractor, it requires to store and manipulate the kernel matrix in the training stage, the size of which is square of the number of samples. When the sample numbers become large, the calculation of eigenvalues and eigenvectors will be time-consuming. In order to enhance the extraction efficiency, an improved KCCA approach is proposed which is based eigenvector set. First, the feature selection method is used to select a subset of samples which are mapped to RKHS, and then the KCCA is used for feature extraction. By doing so, the computational complexity of KCCA is greatly reduced from O（n^3） to O（nL^2） （L〈〈n）. Finally the framework of KCCA plus SVDD-based classifier is used in face recognition. The experimental results demonstrate the proposed method largely reduces the training time and the system storage without deteriorating the recognition accuracy.
出处
《计算机工程与设计》
CSCD
北大核心
2009年第5期1183-1185,1188,共4页
Computer Engineering and Design
基金
甘肃省自然科学基金项目(2007GS04782)。
关键词
人脸识别
核典型相关分析
特征向量选择
支持向量数据描述
特征提取
face recognition
kernel canonical correlation analysis
feature vector selection
support vectors data description
feature extraction
