SynMorph: Generating Synthetic Face Morphing Dataset with Mated Samples

TL;DR

SynMorph addresses the scarcity of public, large-scale face morphing datasets by generating a high-quality synthetic morphing corpus with 2450 identities and over 100k morphs at 1024×1024 resolution, designed to support both S-MAD and D-MAD. The method combines StyleGAN2-based base sample generation with latent-space neutralization, three mated-sample editing strategies, and dual morphing pipelines (GAN-based and landmark-based). Extensive evaluation assesses face image quality via FIQA, morphing vulnerability via MAP, and MAD performance under three protocols across multiple algorithms, showing competitive quality and stronger attack potential in synthetic data while highlighting cross-domain generalization challenges. The dataset and protocols enable scalable benchmarking and training for MAD systems, with practical implications for privacy-friendly, reproducible MAD research and deployment. Future work should further bridge synthetic/non-synthetic gaps and refine guidelines for reporting synthetic data usage in MAD evaluation.

Abstract

Face morphing attack detection (MAD) algorithms have become essential to overcome the vulnerability of face recognition systems. To solve the lack of large-scale and public-available datasets due to privacy concerns and restrictions, in this work we propose a new method to generate a synthetic face morphing dataset with 2450 identities and more than 100k morphs. The proposed synthetic face morphing dataset is unique for its high-quality samples, different types of morphing algorithms, and the generalization for both single and differential morphing attack detection algorithms. For experiments, we apply face image quality assessment and vulnerability analysis to evaluate the proposed synthetic face morphing dataset from the perspective of biometric sample quality and morphing attack potential on face recognition systems. The results are benchmarked with an existing SOTA synthetic dataset and a representative non-synthetic and indicate improvement compared with the SOTA. Additionally, we design different protocols and study the applicability of using the proposed synthetic dataset on training morphing attack detection algorithms.

Figures (15)

• Figure 1: Overall summary and comparison of existing approaches and proposed SynMorph approach for generating synthetic morph dataset.
• Figure 2: Overview of the generation of SynMorph dataset.
• Figure 3: Overview of the generation of SynMorph dataset. Each triplet of images is selected based on SER-FIQ quality score: left-lowest, middle-median, right-highest. In D-MAD cases, IFGS images will be used as non-synthetic enrollment images, IFGD or FRPCA will be used as probe images with wilder capturing conditions.
• Figure 4: Distribution of FaceQnet face image quality scores of non-morphed images.
• Figure 5: Distribution of FaceQnet image quality scores of morphed images.