基于多尺度梯度角和SVM的正面人脸识别方法  

Frontal face recognition method based on multi-scale gradient angular feature and support vector machine

作　　者：赵武锋[1] 严晓浪[1] 

机构地区：[1]浙江大学超大规模集成电路设计研究所

出　　处：《浙江大学学报（工学版）》2008年第4期590-592,617,共4页Journal of Zhejiang University：Engineering Science

摘　　要：为了提高人脸识别算法性能,提出了一种多尺度梯度角(MSGA)和支持向量机(SVM)相结合的新的正面人脸识别方法.分析了梯度角对光照的不敏感特性和反对称双正交小波(ASBW)的导数特性.获取多尺度梯度角特征,并利用其所具有的降噪能力和有效降低表情变化、光照变化等因素引起的影响,使算法具备较强的鲁棒性.采用了分类性能优越的支持向量机技术,提高了泛化能力.并在Yale人脸数据库上与归一化原始数据、小波处理后数据进行了仿真比较,实验数据显示,不论使用主分量分析(PCA)还是线性鉴别分析(LDA)降维,在相同的维数条件下,新方法的识别性能都优于其他方法.A new frontal face recognition method combining multi scale gradient angular （MSGA） feature and support vector machine （SVM） was proposed in order to improve the performance of the face recognition algorithm. Based on gradient＇s insensitiveness to the illumination variation and the derivative feature of anti-symmetrical biorthogonal wavelet （ASBW）, multi-scale gradient angle features were obtained. This MSGA based method is robust due to making use of the ability to eliminate the influences arising from noise, expression change and other factors, while SVM promotes the generalization ability. The recognition results on Yale face database were compared with the normalized original data and those of the wavelet transformation. Whether principal component analysis （PCA） or linear discriminant analysis （LDA） was used for dimensionality reduction, this method outperformed the others under the same dimension.

关 键 词：反对称双正交小波 支持向量机 线性鉴别分析 主成分分析 多尺度梯度角 非负矩阵分解 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]