基于交叉融合特征的人耳识别  

Ear Recognition Based on Intercrossed Feature

作　　者：徐正光[1] 申思[1] 

机构地区：[1]北京科技大学信息工程学院,北京100083

出　　处：《计算机工程》2007年第18期214-216,共3页Computer Engineering

基　　金：国家自然科学基金资助项目(60375002)

摘　　要：结合人耳图像特点和两种整体统计特征提取方法的优缺点,该文用主成分分析(PCA)方法提取图像的表示信息特征,用压缩后的类平均向量中的判别信息获得先验类别特征并根据特征分量的类间类内方差比准则将两种特征交叉融合成新的特征向量。分别在2个不同的人耳图像库中进行识别实验,结果表明,该文提出的交叉融合特征识别方法比传统的PCA和PCA+LDA方法的正确识别率高,而且在有一定程度的光照变化和一定角度变化的情况下仍可获得很好的识别效果。A new method to get intercrossed features according to the ratio between class and within class variance of feature vectors＇ each component is proposed. The intercrossed features are composed of two kinds of statistical features： PCA and compressed discriminant information among mean vectors of different classes. The method can make use of both the descriptive information and classificatory information. The method is evaluated by the recognition rates over two ear image databases. Experiment results show that the method outperforms traditional PCA or PCA and LDA methods. The method is also proved to work well under some variations in lighting and position.

关 键 词：人耳识别 特征提取 主成分分析 特征融合 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]