Researchers have developed CPG-PAD, a novel framework designed to improve the generalization capabilities of presentation attack detection (PAD) models in face recognition systems. This new approach integrates concept guidance into the prompt learning process, using explainable AI (XAI) techniques to identify PAD-relevant visual concepts and generate heatmaps for localized guidance. By incorporating these concepts into the prompt space, CPG-PAD aims to capture generalizable attack cues and reduce overfitting to domain-specific artifacts. Experiments across nine datasets show that CPG-PAD achieves state-of-the-art cross-domain performance. AI
IMPACT This framework could improve the robustness and security of face recognition systems against sophisticated spoofing attacks.
RANK_REASON The cluster contains a research paper published on arXiv detailing a new technical framework. [lever_c_demoted from research: ic=1 ai=1.0]
- arXiv
- CPG-PAD
- Presentation attack detection for face recognition using light field camera
- Prompt-based Concept Injection
- vision-language model
- Visual Concept-driven Enhancement
- Visual-Prompt Decoder
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
