Fingerprint Liveness Detection

What's implemented

	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.

Results

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.

Links

LivDet 2015 Fingerprint Database

The group project

References

LivDet 2015 Fingerprint Liveness Detection Competition

Review of the LivDet Competition Series: 2009 to 2015

Evaluating software-based fingerprint liveness detection using Convolutional Networks and Local Binary Patterns

D. Maltoni, D. Maio, A. Jain, and S. Prabhakar. Handbook of Fingerprint Recognition. Springer Publishing Company, 2009.