A variety of tools for the S/D-MAD tasks.
All tools are written in Python language.
- Reference paper: Deep Face Representations for Differential Morphing Attack Detection (TIFS 2020)
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.
- 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()
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 |
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 |