Reproduce “Conditional Identity Disentanglement for Differential Face Morph Detection” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Sudipta Banerjee, Arun Ross
Researchers at Michigan State University developed a novel differential face morph detection method using a Conditional Generative Adversarial Network (cGAN) for implicit identity disentanglement. The approach achieved superior generalization across different morphing attack types, outperforming existing state-of-the-art methods by up to 13.9% BPCER in cross-dataset evaluations, and demonstrated the ability to recover information about the second subject in a morphed image.
Key insights
- Differential face morph detection can be effectively framed as an information-theoretic problem, where the 'uncertainty' or 'disparity' in disentangling identities reveals morphing.
- A cGAN can implicitly learn to disentangle identities from a potentially morphed image conditioned on a reference image, without requiring explicit morph/non-morph labels during its training.
- The disentanglement process has a significantly greater impact on morphed images, shifting their biometric similarity scores with the reference image distinctly lower, which facilitates robust detection.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.