Research Article

Thresholded Two-Phase Test Sample Representation for Outlier Rejection in Biological Recognition

Table 1

Minimum overall classification error rate and maximum ROC difference for T-TPSR, T-GR, T-PCA, and T-LDA methods, respectively, on the Feret database.

MethodsT-TPTSRT-GRT-PCA(150)T-LDA(149)

ERRopt (%)20.423.230.030.0
(%)33.032.811.91.24

T-PCA(150) indicate that the T-PCA used 150 transform axes for feature extraction, and T-LDA(119) means that the T-LDA used 119 transform axes for feature extraction. Tables 2 and 3 show the method and number of transform axes used in the same way.