Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

SynID: Passport Synthetic Dataset for Presentation Attack Detection

Juan E. Tapia, Fabian Stockhardt, Lázaro Janier González-Soler, Christoph Busch

TL;DR

The paper tackles the scarcity of realistic, ICAO-compliant data for passport Presentation Attack Detection (PAD) in remote verification. It introduces SynID, a hybrid synthetic dataset generated from open data and synthetic faces to produce ICAO-compliant passport images, implemented through a five-component pipeline (template normalization, metadata generation, biometric data selection, layer-based compositing, and pattern/logo processing). The contributions include one of the first ICAO-aligned passport datasets produced in hybrid mode, with reproducibility on request, and comprehensive experiments showing the dataset challenges for state-of-the-art PAD methods, while highlighting Swin Transformer-based models as particularly effective. The work enables privacy-preserving, scalable PAD training and provides a realistic benchmark for cross-country generalization and attack realism, with plans to expand to more countries and attack types in future work.

Abstract

The demand for Presentation Attack Detection (PAD) to identify fraudulent ID documents in remote verification systems has significantly risen in recent years. This increase is driven by several factors, including the rise of remote work, online purchasing, migration, and advancements in synthetic images. Additionally, we have noticed a surge in the number of attacks aimed at the enrolment process. Training a PAD to detect fake ID documents is very challenging because of the limited number of ID documents available due to privacy concerns. This work proposes a new passport dataset generated from a hybrid method that combines synthetic data and open-access information using the ICAO requirement to obtain realistic training and testing images.

SynID: Passport Synthetic Dataset for Presentation Attack Detection

TL;DR

The paper tackles the scarcity of realistic, ICAO-compliant data for passport Presentation Attack Detection (PAD) in remote verification. It introduces SynID, a hybrid synthetic dataset generated from open data and synthetic faces to produce ICAO-compliant passport images, implemented through a five-component pipeline (template normalization, metadata generation, biometric data selection, layer-based compositing, and pattern/logo processing). The contributions include one of the first ICAO-aligned passport datasets produced in hybrid mode, with reproducibility on request, and comprehensive experiments showing the dataset challenges for state-of-the-art PAD methods, while highlighting Swin Transformer-based models as particularly effective. The work enables privacy-preserving, scalable PAD training and provides a realistic benchmark for cross-country generalization and attack realism, with plans to expand to more countries and attack types in future work.

Abstract

The demand for Presentation Attack Detection (PAD) to identify fraudulent ID documents in remote verification systems has significantly risen in recent years. This increase is driven by several factors, including the rise of remote work, online purchasing, migration, and advancements in synthetic images. Additionally, we have noticed a surge in the number of attacks aimed at the enrolment process. Training a PAD to detect fake ID documents is very challenging because of the limited number of ID documents available due to privacy concerns. This work proposes a new passport dataset generated from a hybrid method that combines synthetic data and open-access information using the ICAO requirement to obtain realistic training and testing images.
Paper Structure (14 sections, 2 equations, 8 figures, 1 table)

This paper contains 14 sections, 2 equations, 8 figures, 1 table.

Table of Contents

  1. Introduction
  2. Related Work
  3. Proposed Method
  4. ID-Template Recovery and Normalisation
  5. Subject Metadata Generation
  6. Biometric Data Selection and Filtering
  7. Layer Compositing Pipeline
  8. Pattern and Logo Processing
  9. Metrics
  10. Experiment and Results
  11. Experiment 1 - Binary classifier
  12. Experiment 2 - Leave-one-out classifier
  13. Benchmark with State of the art
  14. Conclusion

Figures (8)

  • Figure 1: Detailed images with all the changes highlighted in colours.
  • Figure 2: Example of signature extraction process. Left to Right: Base image from the dataset. Binarise image. Remove background, add to passport.
  • Figure 3: Detailed logo and pattern information created in order to follow the pattern of the Polish passport.
  • Figure 4: Basic steps for fake passport generation. Left to right: Empty template, intermediate state with demographic text, and final composition with biometric image.
  • Figure 5: Example of three fake passport generations for the following countries. Left to right: Spain, Poland, Portugal.
  • ...and 3 more figures