摘要： 提出定点ICA算法结合均一化欧氏距离的人脸识别方法。该方法不仅可以提取出独立主成分，而且收敛速度快，克服了PCA算法提取特征不独立和一般ICA算法收敛速度慢的缺点。在特征聚类方法中，均一化欧氏距离的聚类效果优于欧氏距离和马氏距离。在正交化预处理过程中，投影特征空间维数的选取根据具体数据库和训练样本集通过试验确定。试验结果表明,该人脸识别方法具有较好的识别能力和鲁 棒性。

关键词: 人脸识别, ICA, 图像特征提取, 均一化欧氏距离

Abstract: A new Face Recognition method base on Fixed-Point algorithm and mean- Euclidean distance is presented. The method can get independent components and convergences very fast,while PCA cannt get independent components and general ICA convergences very slowly.In feature classification, mean- Euclidean distance performances better than general Euclidean distance and Mahalanobis distance. During orthogonalization pre-transaction,the number of dimensionality of Face Space depends on specific face-data and specific samples.Experiments demonstates a good recognition capability and robustness of our system .

中图分类号: 

• TP391.4