Table 1.

Comparisons of proposed and previous research on finger-based recognition.

Category		Methods	Advantage	Disadvantage
Single modal based	Fingerprint	SVM-based quality estimation [1] and minutiae triplets [2]	Cost and size of system are most effective	- Vulnerable to fake attack - Affected by the skin condition of finger
	Finger knuckle print	Subspace [4], Local and global feature [5], Riesz transform [6] and Band-Limited Phase-Only Correlation (BLPOC) [5,9,10,31]	Less affected by the skin condition of finger than fingerprint recognition	More vulnerable to finger movement and skin deformations than fingerprint recognition
	Finger-vein	Radon transform [11], mean curvature [12], maximum curvature [13], Gabor filter [18,20], Hessian filter [19], fuzzy system [21], and convolutional neural network (CNN) [32,33]	- More resistant to fake attacks than fingerprint and finger knuckle-print recognition - Not affected by the skin condition of finger	Affected by shadows caused by NIR light, finger misalignment, and skin light scattering blur
	Finger shape	Wavelet transform [22], and Fourier descriptor and principal component analysis [23]	Not affected by the skin condition of finger	- Affected by thickness of finger according to age or health condition - The device size is bigger than fingerprint, finger knuckle-print, and finger-vein recognition devices - Extraction is hindered by stuck fingers
Multi-modal based	Multiple sensors based	Fusion of fingerprint and finger-vein [24], Fusion of finger-vein, finger shape, fingerprint, and finger knuckle print [25], fusion of finger-vein, fingerprint, and finger knuckle print [26], and fusion of finger-vein and finger knuckle print [27]	Better recognition performance than single-model methods by using 2 or more biometric traits	- High cost and large system size due to the use of 2 or more image-acquisition devices - Slow image-acquisition speed due to inability to acquire multimodal images simultaneously
	Single-sensor based	Handcrafted features and SVM [28,29]	Simultaneous finger-vein, fingerprint, and finger shape recognition using 1 device [28] Simultaneous finger-vein and finger shape recognition using 1 device [29]	Limited recognition performance improvement due to the use of handcraft features
		Deep features by CNN (proposed method)	- Simultaneous finger-vein and finger shape re cognition with 1 device - High recognition performance through the use of deep features	Requires intensive CNN training