Finger-Vein Recognition Using Bidirectional Feature Extraction and Transfer Learning

Table 6

The recognition accuracy and time consumption of different fusion models.

Database	Based network	Feature concatenation		Score fusion
		Accuracy (%)	Time (s)	Score	Accuracy (%)	Time (s)
A and B FV-USM	CNN (VGG19)	98.00	16.88	8 : 2	96.67	13.24
				7 : 3	97.73	13.95
	CNN (ResNet50)	99.67	38.42	9 : 1	99.33	41.48
				6 : 4	99.67	81.71
A and B FV-SIPL	CNN (VGG19)	99.07	21.84	8 : 2	96.06	17.28
				5 : 5	99.54	19.50
	CNN (ResNet50)	99.31	43.70	9 : 1	98.38	46.36
				1 : 9	99.07	229.77