From Vision to Text: A Compact Multimodal Approach for Robust, Cross-Domain Presentation Attack Detection on ID Cards

2026-06-05

AI Summary

This work addresses the challenge of cross-domain presentation attack detection (PAD) for identity documents, where privacy constraints limit the availability of real-world data and hinder model generalization. To overcome this, the authors propose a compact multimodal framework that, for the first time, jointly leverages visual and textual information for document PAD. The approach integrates generative and discriminative components and enhances cross-domain generalization through supervised fine-tuning. Experimental results demonstrate strong performance on both real and synthetic datasets, highlighting the critical roles of model capacity and diverse real data in achieving robust PAD. Furthermore, the study reveals significant limitations of existing synthetic datasets under zero-shot evaluation settings, casting doubt on their reliability as benchmarks for realistic PAD assessment.

Abstract

Cross-domain shifts challenge Presentation Attack Detection (PAD) on ID Cards, given the restricted data available due to privacy concerns. This work proposes a compact multimodal model, based on new generative and discriminative blocks, which combines visual and textual data for PAD on genuine and synthetic ID images. While multimodal models exhibit strong generalisation after supervised fine-tuning, they fail in zero-shot settings. Our findings underscore that model capacity and real-world data are essential for reliable PAD, while existing synthetic datasets may not reflect real-world challenges. We argue for a re-evaluation of synthetic data as a benchmark and emphasise the need for more realistic, diverse datasets to advance PAD research.

Problem

• Presentation Attack Detection
• Cross-domain
• ID Cards
• Synthetic Data
• Multimodal

Innovation

• multimodal learning
• presentation attack detection
• cross-domain generalization
• synthetic data evaluation
• ID card security