Generating Automatically Print/Scan Textures for Morphing Attack Detection Applications

TL;DR

This work tackles the data scarcity challenge in Morphing Attack Detection (MAD) by proposing two automated data-generation strategies to simulate printed-and-scanned face images: (i) GAN-based transfer-style methods (Pix2pix and CycleGAN) to convert digital bona fide/morphed faces into realistic print/scan counterparts, and (ii) a semi-automatic handcrafted texture-transfer method that injects hardware-induced print textures using a 50-color palette. Evaluations on FRGCv2/FERET show that augmenting training with synthetic print/scan data reduces detection error, achieving EER values as low as 3.84% and 1.92% in certain setups. The study uses Frechet Inception Distance (FID) to assess realism and MACER/BPCER with a Leave-One-Out SVM framework to measure MAD performance, demonstrating that frequency-domain features (DCT, SRM) are particularly effective for print/scan morphing detection. Overall, the results indicate that realistic synthetic data can substantially improve MAD robustness and generalization to diverse morphing tools and devices, paving the way for larger, more representative training datasets in operational passport scenarios.

Abstract

Morphing Attack Detection (MAD) is a relevant topic that aims to detect attempts by unauthorised individuals to access a "valid" identity. One of the main scenarios is printing morphed images and submitting the respective print in a passport application process. Today, small datasets are available to train the MAD algorithm because of privacy concerns and the limitations resulting from the effort associated with the printing and scanning of images at large numbers. In order to improve the detection capabilities and spot such morphing attacks, it will be necessary to have a larger and more realistic dataset representing the passport application scenario with the diversity of devices and the resulting printed scanned or compressed images. Creating training data representing the diversity of attacks is a very demanding task because the training material is developed manually. This paper proposes two different methods based on transfer-transfer for automatically creating digital print/scan face images and using such images in the training of a Morphing Attack Detection algorithm. Our proposed method can reach an Equal Error Rate (EER) of 3.84% and 1.92% on the FRGC/FERET database when including our synthetic and texture-transfer print/scan with 600 dpi to handcrafted images, respectively.

Figures (10)

• Figure 1: Illustration of our pixel-wise print/scan simulation network Mitkovski.
• Figure 2: Illustration of our CycleGan print/scan simulation network.
• Figure 3: Example of 50 texture palettes with the QR for each colour.
• Figure 4: Example of all the methods used to generate print/scan images from FRGC dataset.
• Figure 5: Example Texture-transfer technique applied to different sources and resolutions. Left to right: Digital image, Bond paper 300 dpi, Bond paper 600 dpi, Glossy paper 300 dpi and Glossy paper 600 dpi.