Private Face Image Generation Method Based on Deidentification in Low Light.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Biometrics

Background:

• Face recognition technology poses privacy risks by linking facial features to personal information.
• Existing algorithms struggle with privacy protection in low-light conditions.
• De-identification is crucial for safeguarding sensitive biometric data.

Purpose of the Study:

• To develop a method for generating private face images under low-light conditions.
• To de-identify facial data while preserving image structure and utility.
• To enhance user privacy against facial recognition systems.

Main Methods:

• Pretraining light enhancement and attenuation networks.
• Enhancing low-light images and intercepting facial areas.
• Generating de-identified latent codes with feature disentanglement.
• Creating private low-light face images using a face generation network.

Main Results:

• The proposed method successfully generates low-light private face images.
• Generated images exhibit high structural similarity to original photos.
• Effectively reduces face recognition accuracy, enhancing privacy.
• Maintains image practicability for various applications.

Conclusions:

• The de-identification method offers effective privacy protection for face images.
• Achieves superior performance in generating private low-light facial images compared to state-of-the-art methods.
• Balances privacy preservation with image usability in challenging lighting conditions.