gdubrg / MAD-Tools

A variety of tools for the S/D-MAD tasks

MAD Tools

A variety of tools for the S/D-MAD tasks.
All tools are written in Python language.

Differential Morphing Attack Detection

• Reference paper: Deep Face Representations for Differential Morphing Attack Detection (TIFS 2020)

Image pre-processing

The D-MAD classifier is trained on embeddings extracted with the DNN presented in

You can extract features through the extract_features.py script available in this repo. Follow inside comments for an easy use.
Before, you have to install the following packages:

• Mxnet (tested version 1.4.0 running on CPU)
• OpenCV 4.4.0

and other minor packages depending on your original setup (like tqdmm, ...).

Click here to download the ArcFace parameters. For simplicity, put the file in the feature_extraction directory.

Classifier

• SVM with rbf kernel
• Machine learning tool: scikit-learn 0.23.2
• Trained on PMDB dataset; data not balanced (280 genuine, 1108 impostor)
• Features MUST be subtracted! (probe - reference)
• Load the classifier file through the pickle package. Example:

import pickle
with open(<path>, 'r') as f:
    classifier = pickle.load(f)
...
classifier.predict()

Digital images

Train dataset	Train-Test Images	Alpha	Couples with	EER on MorphDB	Model
PMDB	Digital-Digital	0.55	Criminal	0.0%	link
PMDB	Digital-Digital	0.55	Accomplice	0.0%	link
PMDB	Digital-Digital	0.55	Both	0.0%	link
PMDB + MorphDB	Digital-Digital	0.55	Both	-	link

Printed and Scanned images

Train dataset	Train-Test Images	Alpha	Couples with	EER on MorphDB	Model
PMDB	P&S-P&S	0.55	Criminal	0.0%	link
PMDB	P&S-P&S	0.55	Accomplice	0.0%	link
PMDB	P&S-P&S	0.55	Both	0.0%	link
PMDB + MorphDB	P&S-P&S	0.55	Both	-	link