摘要
针对人脸识别中的小样本问题,提出一种快速核主元分析(FKPCA)与双决策子空间的人脸识别方法。利用FKPCA方法将原始样本空间映射到高维空间,在高维空间中实现原始样本的降维,在双决策子空间分别用Fisher准则和类间散布判决准则提取常规信息和非常规信息,通过加权欧式距离进行信息融合并用最近邻分类器进行识别。在ORL人脸库上的实验结果表明,该方法具有较高的识别率与较快的识别速度。
Aiming at the Small Sample Size（3S） problem in face recognition, this paper presents face recognition method based on Fast Kernel Principle Component Analysis（FKPCA） and double decision subspace. FKPCA method is used to map original input space to high-dimensional space and reduce the dimension of input samples. It can get the regular decision information by using Fisher criterion and gain the irregular decision information by employing between-class scatter criterion. Weighted Euclidean distance is employed to fuse two kinds of features, and classification is implemented with nearest neighbor classifier. Experimental results on ORL database show that this method can reach a higher correct recognition rate and good speed.
出处
《计算机工程》
CAS
CSCD
北大核心
2010年第18期182-184,共3页
Computer Engineering
基金
国家自然科学基金资助项目(60673190)
江苏大学高级专业人才科研启动基金资助项目(05JDG020)
关键词
快速核主元分析
双决策子空间
特征融合
加权欧式距离
Fast Kernel Principle Component Analysis（FKPCA）
double decision subspace
feature fusion
weighted Euclidean distance
