Shawn-Shan/fawkes
Fawkes
Fawkes is an adversarial image generator and facial recognition cloaking tool designed to protect privacy by obfuscating facial features in photos. It functions as an image privacy obfuscator that adds invisible pixel perturbations to images, preventing facial recognition models from accurately identifying a person while keeping the image visually clear to humans.
The system employs adversarial perturbation mapping and feature-space obfuscation to mislead machine learning classifiers. By utilizing an iterative optimization loop and model-agnostic noise generation, it modifies facial representations to prevent recognition systems from extracting a consistent identity across different architectures.
Features
- Biometric Evasion - Applies invisible pixel changes to photos to prevent facial recognition models from accurately identifying individuals.
- Adversarial Image Perturbations - Implements mathematically calculated pixel perturbations to mislead facial recognition models without altering human perception.
- Adversarial Image Manipulations - Crafts adversarial images to exploit facial recognition model behavior while keeping the image clear to humans.
- Facial Recognition - Blocks facial recognition systems from identifying individuals by adding invisible perturbations to images.
- Recognition Cloaking - Provides a tool that adds invisible pixel perturbations to prevent facial recognition models from identifying a person.
- Identity Cloaking - Modifies facial feature representations within images to prevent recognition systems from extracting a consistent identity.
- Image Privacy Obfuscation - Protects facial features in photos by altering pixels to mislead machine learning classifiers.
- Latent Space Manipulations - Alters the latent representations of faces to deviate from the original biometric signature.
- Model-Agnostic Noise - Produces image perturbations that generalize across different facial recognition architectures.
- Adversarial Optimization Loops - Uses iterative processes to refine adversarial noise through repeated gradient updates.
- Computer Vision - Privacy tool for obfuscating faces against recognition.
- Computer Vision and Image Processing - Privacy tool for obfuscating faces from recognition.
- Data Cloaking Tools - Tool to protect against facial recognition systems.
