Finger-Vein Recognition Using Bidirectional Feature Extraction and Transfer Learning

Table 7

Comparison of recognition rates and time consumption of finger-vein recognition experiments under different feature extraction algorithms.

Paper	Publication year	Feature extraction method	Database	Accuracy rate (%)	Time (s)
Qui et al. [36]	2016	Dual-sliding window localization + pseudoelliptical transformer + 2D-PCA	FV-USM	97.02	—
He and Chen [37]	2018	Gabor + uniform + LBP + DBN	FV-USM	97.93	229.23
		Improved Gabor + uniform + LBP + DBN		98.21	235.23
		CNN (max pooling + improved activation function)		96.43	—
Ding [38]	2019	Double linear Weber local descriptor	FV-USM	99.25	157.74
Das et al. [39]	2019	CNN (proposed CNN)	FV-USM	97.48	—
Yuan [40]	2020	Multiscale LBP + DBN	FV-USM	99.59	339.40
In this paper	2021	CNN (VGG19) + feature concatenation	A and B FV-USM	98.00	16.88
		CNN (ResNet50) + feature concatenation		99.67	38.42
		CNN (VGG19) + feature concatenation	A and B FV-SIPL	99.07	21.84
		CNN (ResNet50) + feature concatenation		99.31	43.70