摘要
提出了一种新的基于差空间的最大散度差鉴别特征抽取方法。该方法首先通过构造人脸图像的差空间,部分地消除由于光照条件不同而引起的人脸图像的不稳定性,然后采用最大散度差鉴别准则函数进行最优鉴别特征的抽取,这样从根本上避免了传统的Fisher线性鉴别分析中存在的“小样本问题”。最后,在ORL标准人脸库和Yale人脸库上的实验结果验证了本文算法的有效性。
A new method of discriminant feature extraction based on scatter difference criterion in residual space was developed in this paper. Firstly, the instability of face images due to some different illuminations was moderated by constructing the residual space for face images. Then, the maximum scatter difference criterion function was adopted to extract a set of optimal features. As a result, the small size sample problem suffered from the traditional Fisher linear discriminant analysis was utterly avoided. Finally, extensive experiments performed on both 0RL face database and Yale face database verify the effectiveness of the proposed method.
出处
《计算机应用》
CSCD
北大核心
2006年第10期2460-2462,2465,共4页
journal of Computer Applications
基金
江苏省高校自然科学基金资助项目(05KJB520152)
江苏省博士后科研资助计划项目
关键词
差空间
最大散度差鉴别分析
人脸识别
residual face space
maximum scatter difference discriminant analysis
face recognition
