Verification of Dynamic Holographic Behavior in Identity Documents

2026-07-08 · Source: cs.CV updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Cybersecurity & Data Privacy · Depth: Expert, extended

Summary

A new study introduces MIDV-DynAttack, a public dataset designed to improve remote verification of Optically Variable Devices (OVDs, or holograms) on identity documents. This dataset extends the existing MIDV-Holo with 1,200 new attack videos, tripling the number of attack samples, specifically focusing on realistic static and dynamic fraud scenarios. Traditional automated verification methods struggle with these dynamic attacks and unseen forgeries due to limited public data. The researchers also propose HoloVerif, a novel verification method that analyzes OVD dynamic behavior and appearance. HoloVerif can be trained without dynamic attack samples and achieves new leading performance, particularly against dynamic attacks. A benchmark of existing approaches confirms their inability to effectively counter these challenging dynamic frauds. Code and dataset are publicly available.

Key takeaway

For AI Security Engineers developing identity document verification systems, you should prioritize solutions capable of detecting dynamic holographic attacks. Integrate HoloVerif's background estimation and pseudo-labeling approach to model OVD temporal dynamics, as existing methods are vulnerable to sophisticated forgeries. Additionally, combine Model Verification with Presentation Attack Detection to counter complex threats like document swapping and photo replacement, ensuring comprehensive security against evolving fraud techniques.

Key insights

Automated OVD verification requires dynamic behavior analysis and datasets with unseen, realistic attack scenarios.

Principles

• Background subtraction enhances faint holographic signals.
• Pseudo-labeling can train classifiers for complex OVD behaviors.
• Augmentations improve robustness against dynamic attacks.

Method

HoloVerif preprocesses OVD video sequences via background subtraction and HSV-based normalization, then uses a pseudo-labeled frame classifier, followed by a threshold-based decision stage on valid frame ratios.

In practice

• Integrate background subtraction to isolate OVD signals.
• Use pseudo-labeling for training OVD behavior models.
• Combine Model Verification with Attack Detection for robust systems.

Topics

• Optically Variable Devices
• Identity Document Verification
• Hologram Authenticity
• Presentation Attack Detection
• MIDV-DynAttack Dataset
• HoloVerif Method

Code references

• EPITAResearchLab/pouliquen.25.icdar

Best for: AI Engineer, Computer Vision Engineer, Research Scientist, AI Scientist, Machine Learning Engineer, AI Security Engineer

Related on AIssential

• Temporal Modeling of Optically Variable Devices in Identity Documents — Modeling temporal dynamics of OVDs with self-supervised methods defeats advanced identity document fraud without attack samples.
• From Forgeries to Foundation Models: A Systematic Survey of Identity Document Attack and Detection — Generative AI has fundamentally shifted identity document forgery, rendering current detection benchmarks inadequate against advanced attacks.
• CPG-PAD: Concept-Informed Prompts Guided Presentation Attack Detection — CPG-PAD enhances presentation attack detection generalization by integrating XAI-derived visual concepts into VLM prompt learning.
• Efficient, Robust, and Anti-Collusion Fingerprinting of Image Diffusion Models — A novel T2I model fingerprinting method uses a Personalized Normalization Module and anti-collusion transformations to ensure robustness against malicious model combinations.
• Microsoft's research argues AI media authentication doesn't work reliably, yet new laws assume it does — Current AI media authentication methods are unreliable, requiring combined approaches and facing legislative overreach.
• From Vision to Text: A Compact Multimodal Approach for Robust, Cross-Domain Presentation Attack Detection on ID Cards — Compact multimodal PAD models need real-world data and fine-tuning for robust cross-domain performance, challenging synthetic dataset utility.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.CV updates on arXiv.org.