Researchers have developed a novel approach for zero-shot domain generalization in presentation attack detection (PAD) for ID cards. This method utilizes a Prototypical Network with an EfficientNet-V2-b0 backbone, requiring only four genuine samples per class to establish reliable prototypes. An episodic training regime allows the network to learn universal attack cues by varying card domains while keeping PAD classes constant. The system achieved an average Equal Error Rate of approximately 9% on a multi-country dataset and the DLC-2021 benchmark, outperforming existing baselines and enabling privacy-preserving, scalable cross-jurisdictional remote onboarding. AI
IMPACT This research could improve the security and efficiency of remote onboarding processes by enabling more accurate and privacy-preserving ID verification.
RANK_REASON The cluster contains an academic paper detailing a new method for presentation attack detection.
- DLC-2021
- EfficientNet-V2-b0
- Prototypical Networks for Few-shot Learning
- Hugging Face
- ID-Card
- Presentation attack detection for face recognition using light field camera
AI-generated summary · Google Gemini · from 2 sources. How we write summaries →