摘要
本文通过研究光照下人脸图像,提出了一种新的光照不变量的特征提取方法,我们称为特征脸。不同于Eigenfaecs和Fisherfaces,本文中的特征脸是一个光照不变量。我们给出了理论的证明,确保特征脸具有光照不变性。提议的方法首先通过梯度变换,将图像从像素域变换到梯度域来处理,然后从梯度域提取光照不变性的特征用于人脸识别,实验结果表明,该方法能有效地提高人脸识别对光照的鲁棒性,同时该方法非常简单,能够用于实时人脸识别系统。
A novel method to extract illumination invariant is proposed for face recognition under varying lighting conditions called Featurefaces.Different from Eigenfaces and Fisherfaces,Featurefaces is an illumination invariant. The theoretical justification and analysis are provided for Featurefaces,which can ensure that Featurefaces is an illumination invariant.In the proposed approach,a gradient transform is employed to map pixel domain into gradient domain for analysis.Then Featurefaces is extracted from the gradient domain for face recognition.In a way, Featurefaces is effective method for face recognition with or without illumination variation.The very high recognition rates achieved by Featurefaces using PCA recognition on Yale B have shown that Featurefaces is an illumination invariant.In addition,the proposed method is very simple and easy to implement.
基金
高等学校博士学科点专项科研基金(20060611009)
关键词
人工智能
光照不变量
特征脸
人脸识别
artificial intelligence
illumination invariant
feature faces
face recognition
