AURA (Adversarial Use for Reliable Assessment) is a research framework for generating and benchmarking architecture-agnostic adversarial perturbations against OCR systems while preserving human readability. Inspired by human visual cognition, Gestalt principles, and perceptual illusions, AURA evaluates structural, geometric, photometric, and ...
ocr computer-vision deep-learning optical-character-recognition evaluation-framework adversarial-machine-learning human-vision adversarial-examples adversarial-attacks machine-learning-security ai-security visual-perception robust-ai llm-security perturbation-generation
-
Updated
Jun 25, 2026 - Python