摘要
横向比较特征提取方法,综合考虑认证率和特征提取时间两方面因素,该文认为特征脸结合线性判别分析方法是研究的4种特征提取方法中最优的方法。通过对投影空间维数的研究,最佳投影空间维数同数据库本身类内图像的相似程度和每一类的样本数目同方向增长,它们之间存在定性关系而非定量关系。相似度匹配方法的研究结果表明余弦距离分类器分类效果最佳。
Eigenfaces combining linear discriminant analysis method is taken as the best by comparing different methods, considering synthetically verification rate and feature extraction time. Through study of projection space dimensions, it is indicated that the best projection dimension is connected with the similarity between images of one kind of the data base, with the number of samples of one kind. But their relation is qualitative, not quantitative, Studying of similarity match methods, it shows that cosine distance is best for classification.
出处
《计算机工程》
EI
CAS
CSCD
北大核心
2006年第11期225-227,247,共4页
Computer Engineering
关键词
人脸识别
特征提取
距离分类器
线性判别分析
Face recognition
Feature extraction
Distance-classifier
Linear discriminant analysis （LDA）
