Table 3: Comparisons of the recognition accuracy (%) of different algorithms with different numbers of training sample per class (PolyU).

	100 Classes			150 Classes			200 Classes
Training Samples Per Class	3	5	7	3	5	7	3	5	7
IR-DDBN	99.44	99.52	100	99.25	99.55	99.98	98.75	99.39	99.75
CNN	98.14	99.10	100	97.40	98.60	99.11	96.14	97.50	98.67
PCA	96.71	98.80	99.67	97.81	99.20	99.78	97.21	98.9.	99.67
HOG	93.00	97.20	99.67	91.62	96.13	98.67	91.50	95.80	98.00
LBP	91.00	93.80	97.33	87.14	91.33	95.56	87.86	91.4	95.17