| Filename | Description | |
|---|---|---|
| Feature extraction | extract_features.py |
Convnet-features (CNN) |
| Data augmentation | augment_data.py |
Horizontal flip, crop 5+5 patches |
| Models | convnet_svm.py |
SVM |
convnet_nn.py |
Neural network | |
imagenet_finetune.py |
Inception v3 |
CNN feature extraction requires CNN-RFW.
Inception v3 settings: samples_per_epoch=250, nb_epoch=25.
| Pipeline | ACC | ACE |
|---|---|---|
| BSIF + NN | 85.24 | 14.28 |
| AUG + BSIF + SVM | 84.86 | 14.44 |
| AUG + BSIF + NN | 85.93 | 13.82 |
| CNN-RFW + SVM | 81.16 | 17.95 |
| CNN-RFW + NN | 81.84 | 18.16 |
| Inception v3 | 66.60 | 28.93 |
| BSIF/CNN-RFW + NN | 82.85 | 17.15 |
Average classification error: ACE = (FPR + FNR)/2
Inception v3 with ImageNet weights couldn't perform well for our peculiar images of the fingerprints.
BSIF/CNN-RFW means mixed features.
LivDet 2015 Fingerprint Database
LivDet 2015 Fingerprint Liveness Detection Competition
Review of the LivDet Competition Series: 2009 to 2015
D. Maltoni, D. Maio, A. Jain, and S. Prabhakar. Handbook of Fingerprint Recognition. Springer Publishing Company, 2009.