Face Re-morphing detects morph attacks using similarity changes

Face morphing attacks pose a critical threat to face recognition systems, as a single morphed ID document can match multiple people. Existing differential morphing attack detection (D-MAD) methods rely on static feature differences, constituent-face reconstruction, or multi-cue fusion. In a new paper accepted to IJCB2026, researchers Jie Jin, Masakatsu Nishigaki, and Tetsushi Ohki propose Face Re-morphing, a fundamentally different approach: they apply an additional morphing operation to the document image and the trusted live image, then use the change in cosine similarity between document–live and live–re-morphed pairs as the detection score.

Experiments on FRLL-Morphs and FEI Morph validate the technique across various morphing conditions, re-morphing methods, and face recognition models. The method delivers favorable results on AMSL and performs particularly well on the Criminal condition of FEI Morph Version 1 when using MorDIFF. The team concludes that re-morphing-induced similarity changes offer a complementary and effective cue for D-MAD, strengthening defenses against morphing attacks in identity verification.