ndido98 / siamese

Official implementation of the paper "A double siamese framework for differential morphing attack detection." (Borghi et al., Sensors 2021)

A double siamese framework for differential morphing attack detection

This is the repository that holds the official reference implementation for the paper "A double siamese framework for differential morphing attack detection" (Borghi et al., 2021).

Requirements

The required packages are present in the requirements.txt file. To install them, run the following command:

pip install -r requirements.txt

Usage

The siamese package exposes a get_prediction function which, in its simplest form, takes in input a document and a live image, and returns a morphing prediction. 0 means that the document image is bona fide, while 1 means that the document image is morphed.

Weights are automatically downloaded during the first call to get_prediction.

from siamese import get_prediction
import cv2 as cv

# Load the document and the live image
document = cv.imread("document.png")
live = cv.imread("live.png")

# Get the prediction
prediction = get_prediction(document, live)

This function also allows the user to specify the device to use for the computation (i.e. CPU or GPU) with the optional device parameter. The default value is cpu.

from siamese import get_prediction
import cv2 as cv

# Load the document and the live image
document = cv.imread("document.png")
live = cv.imread("live.png")

# Get the prediction
prediction = get_prediction(document, live, device="cuda:0")

Finally, the function supports computing batched predictions, by passing two lists of equal length: one containing the documents and the other containing the live images. The function will return a list of predictions.

from siamese import get_prediction
import cv2 as cv

# Load the documents and the live images
documents = [cv.imread("document1.png"), cv.imread("document2.png")]
lives = [cv.imread("live1.png"), cv.imread("live2.png")]

# Get the predictions
predictions = get_prediction(documents, lives, device="cuda:0")

Acknowledgement

When using the code from this repository, please cite the following work:

@article{borghi2021double,
  title={A double siamese framework for differential morphing attack detection},
  author={Borghi, Guido and Pancisi, Emanuele and Ferrara, Matteo and Maltoni, Davide},
  journal={Sensors},
  volume={21},
  number={10},
  pages={3466},
  year={2021},
  publisher={MDPI}
}