Morphing Attack Potential

This repository contains the Python source code to compute the Morphing Attack Potential (MAP) metric introduced in [1].

Overview

MAP.py provides the core functionalities to calculate the MAP metric.

ComputeMAP.py contains a script to compute the MAP metric given the following input parameters:

input_folder_path - the folder containing a text file for each Face Recognition Systems (FRSs) evaluated, containing the corresponding comparison scores;
output_folder_path - the folder path where the MAP values will be saved;
frs_info_file_path - the path of a JSON file containing information about the evaluated FRSs (see FRS information file format section below).

PaperScores is a folder containing the scores used to generate the results reported in the MAP paper useful to test the Python scripts.

FRS score file text format

In order to be loaded by ComputeMAP.py as a FRS score file, a text file must contain for each morphed image a number of lines equal to the number of subjects contributing to the generation process (usually two). Given a morphed image and one of the contributing subject, the corresponding score line must fulfill the following format:

morphID subjectID score1 score2 ... scoreN

where:

morphID is the unique identifier of the given morphed image (e.g., M001, M002, etc.);
subjectID is the unique identifier of the given subject;
score1 score2 ... scoreN are the scores obtained by comparing the morphed image against all probe images of the given subject.

Note that, to compute the MAP metric on multiple FRSs, the morphID and subjectID identifiers must be the same in the different score files.

Please check the score files stored into the PaperScores folder for some FRS score file examples.

FRS information file format

To properly compute the MAP metric, a JSON file containing important information about all evaluated FRSs (frs_info_file_path) needs to be passed to the ComputeMAP.py script.

The information must be stored as a dictionary with an entry for each FRS. Each entry must be of the following format:

key - the FRS name string (the same name of the FRS score file stored into the input_folder_path folder);
item - a list containing the decision threshold of the FRS and a boolean value indicating whether the FRS returns a similarity (true) or a dissimilarity (false) score (e.g., [0.5,true]).

Please check the JSON file stored into the PaperScores folder for an example of FRS information file.

How to run

Clone the repository

git clone https://github.com/MatteoFerrara/Morph-Attack-Potential.git

Execute the ComputeMAP.py script

python3 ComputeMAP <input_folder_path> <output_folder_path> <frs_info_file_path>

PaperScores folder contains the scores used to generate the results reported in the MAP paper. To reproduce the results reported in Tables IV and V execute the following code:

python3 ComputeMAP PaperScores/D/ PaperScores/D/ PaperScores/FRS_Info.json

python3 ComputeMAP PaperScores/PS/ PaperScores/PS/ PaperScores/FRS_Info.json

Citation

Please cite [1] in all publications and works that use this code.

Bibliography

[1] M. Ferrara, A. Franco, D. Maltoni, and C. Busch, "Morphing Attack Potential", in proceedings of the IEEE International Workshop on Biometrics and Forensics (IWBF), Salzburg, Austria, April 2022.