Feature Interaction-Based Face De-Morphing Factor Prediction for Restoring Accomplice's Facial Image

Area of Science:

• Computer Science
• Artificial Intelligence
• Biometrics

Background:

• Face morphing attacks degrade the link between face images and identity, challenging face recognition systems.
• Current detection methods cannot reconstruct accomplice face images.
• Existing deep learning de-morphing techniques often neglect inherent morphing factors.

Purpose of the Study:

• To develop a novel face de-morphing method for restoring accomplice identity information.
• To address the limitations of existing techniques in reconstructing morphed faces.

Main Methods:

• A novel face de-morphing method predicting a de-morphing factor.
• Utilizing a channel-wise attention mechanism for feature interaction.
• Integrating correlation between morphed and reference images for factor prediction.
• Mapping images to StyleGAN latent space and using inverse linear interpolation for restoration.

Main Results:

• The proposed method demonstrates superior performance in restoring accomplice facial images.
• Achieved improved restoration accuracy and image quality compared to existing methods.
• Successfully reconstructs identity information lost in morphing attacks.

Conclusions:

• The novel face de-morphing method effectively restores accomplice identity information.
• The technique offers enhanced accuracy and image quality in face restoration.
• This approach advances the field of face morphing attack countermeasures.