Exploration of Incremental Synthetic Non-Morphed Images for Single Morphing Attack Detection

TL;DR

The paper tackles the challenge of limited bona fide data for S-MAD by injecting synthetic non-morphed face images into training and evaluating cross-dataset generalization. Using FERET, FRGCv2, and SMDD with morphing tools, it demonstrates that a carefully controlled mix of synthetic data improves detection performance, while training purely on synthetic data yields suboptimal results. EfficientNet-B2 and MobileNetV3-large serve as practical backbones, with MACER and BPCER as evaluation metrics and DET/D-EER analyses framing the results. The findings offer practical guidance for deploying S-MAD systems under privacy constraints, highlighting the value of mixed synthetic-real training and the need for further exploration of augmentation strategies and additional backbones.

Abstract

This paper investigates the use of synthetic face data to enhance Single-Morphing Attack Detection (S-MAD), addressing the limitations of availability of large-scale datasets of bona fide images due to privacy concerns. Various morphing tools and cross-dataset evaluation schemes were utilized to conduct this study. An incremental testing protocol was implemented to assess the generalization capabilities as more and more synthetic images were added. The results of the experiments show that generalization can be improved by carefully incorporating a controlled number of synthetic images into existing datasets or by gradually adding bona fide images during training. However, indiscriminate use of synthetic data can lead to sub-optimal performance. Evenmore, the use of only synthetic data (morphed and non-morphed images) achieves the highest Equal Error Rate (EER), which means in operational scenarios the best option is not relying only on synthetic data for S-MAD.

Figures (4)

• Figure 1: Examples of the images from the datasets. In the first row, there are examples from the FERET subset, the second row shows examples from the FRGC subset, and the third row shows the examples of the SMDD dataset. The first column is the bona fide image of the subject 1, the bona fide of the subject 2 is on the last column. Between the subject 1 and subject 2 columns are the morph images: Facefusion FaceFusion-Morph, Face morpherFaceMorpher-Morph, OpenCV Bradski-OpenCV-Morph-2000 and UBO UBO-Morphing, respectively.
• Figure 2: S-MAD workflow, training with FRGCv2 and testing with FERET.
• Figure 3: Examples of the pre-processing step done with MTCNN MTCNN to FERET (a.), FRGCv2 (b.) and SMDD (c.) datasets, respectively. Each image shows side-by-side bona fide and morph.
• Figure 4: DET curves of the models achieved with the test set, first and second columns FRGCv2 was employed as test dataset. The third and fourth columns are the results employing FERET as a test dataset.