Ear Recognition Based on Intercrossed Feature
Si Shen · Jisuanji gongcheng · 2007
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 discriminante 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.