DCMorph: Face Morphing via Dual-Stream Cross-Attention Diffusion

arXiv cs.CV / 4/24/2026

📰 NewsSignals & Early TrendsIdeas & Deep AnalysisModels & Research

共有:

Key Points

The paper introduces DCMorph, a dual-stream diffusion framework for face morphing that targets identity verification systems by conditioning on both identity signals and latent-space representations.
It improves over image-level and GAN-based approaches by using decoupled cross-attention interpolation to inject identity-specific features during denoising and by applying DDIM inversion with spherical interpolation for geometrically consistent latent initialization.
The method enables explicit dual-identity conditioning from two source faces, aiming to avoid the limitations of existing diffusion-based morphing techniques.
Experiments evaluating attacks against four state-of-the-art face recognition systems show DCMorph achieves the highest attack success rates at both tested operating thresholds.
The authors report that DCMorph’s morphing attacks are also difficult to detect with current morphing-attack detection solutions, highlighting a defensive gap.

Abstract

Advancing face morphing attack techniques is crucial to anticipate evolving threats and develop robust defensive mechanisms for identity verification systems. This work introduces DCMorph, a dual-stream diffusion-based morphing framework that simultaneously operates at both identity conditioning and latent space levels. Unlike image-level methods suffering from blending artifacts or GAN-based approaches with limited reconstruction fidelity, DCMorph leverages identity-conditioned latent diffusion models through two mechanisms: (1) decoupled cross-attention interpolation that injects identity-specific features from both source faces into the denoising process, enabling explicit dual-identity conditioning absent in existing diffusion-based methods, and (2) DDIM inversion with spherical interpolation between inverted latent representations from both source faces, providing geometrically consistent initial latent representation that preserves structural attributes. Vulnerability analyses across four state-of-the-art face recognition systems demonstrate that DCMorph achieves the highest attack success rates compared to existing methods at both operational thresholds, while remaining challenging to detect by current morphing attack detection solutions.