A Complete Subspace Analysis of Linear Discriminant Analysis and Its Robust Implementation

Table 3

Comparisons of misclassification rates (%) of several methods on face databases.

Databases	Training size	DLDA	PCA+LDA	MMC	DCV	LDA/QR	LDA/GSVD	Ours
ORL		2.25 (1.68)	2.50 (2.68)	2.50 (1.66)	1.50 (1.74)	3.25 (2.68)	2.50 (2.04)	2.25 (1.84)
		2.00 (2.58)	2.00 (2.83)	2.50 (2.83)	2.00 (2.58)	3.00 (2.35)	2.75 (3.40)	1.25 (1.31)
		2.00 (2.29)	4.75 (2.94)	2.00 (2.58)	5.25 (4.31)	3.75 (2.58)	6.25 (4.75)	2.00 (2.37)
Yale		21.00 (10.24)	16.36 (8.62)	11.51 (6.72)	12.12 (8.33)	20.60 (9.16)	12.12 (9.34)	9.09 (6.22)
		20.00 (6.66)	15.15 (7.93)	27.27 (10.21)	10.90 (8.85)	21.21 (8.85)	10.90 (5.39)	7.87 (5.19)
		19.39 (10.93)	12.72 (8.66)	14.54 (9.34)	9.09 (9.07)	20.00 (11.15)	9.09 (9.07)	8.48 (9.02)