CNN-Based Transfer Learning for 3D Knuckle Recognition

Table 2

Results for pretrained models in one holdout run. Bold means these three models have the highest accuracy.

Models	Sensitivity	Recall	Precession	F1 score
DenseNet201	92.59	92.59	94.56	92.51
Xception	97.35	97.35	97.80	97.26
InceptionV3	99.07	99.07	99.18	99.07
NASNetMobile	92.92	92.92	94.44	92.22
ResNet50	91.80	91.80	92.44	90.61
VGG16	7.34	7.34	10.90	4.60
AlexNet	73.54	73.54	80.74	73.19