Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

DocXPand-25k: a large and diverse benchmark dataset for identity documents analysis

Julien Lerouge, Guillaume Betmont, Thomas Bres, Evgeny Stepankevich, Alexis Bergès

TL;DR

The paper tackles the lack of large, diverse public datasets for identity document analysis by introducing DocXPand-25k, a synthetic yet richly labeled collection of 24,994 IDs drawn from nine vector templates and placed into diverse real backgrounds. It provides comprehensive annotations for classification, localization, barcode/face detection, and OCR text fields, along with an open-source generation toolkit. Perceptual realism is validated via a LPIPS-based comparison to real-ID captures, and baseline experiments demonstrate practical utility with lightweight models for classification and localization, and a Tesseract-based OCR benchmark that remains far from industrial adequacy, indicating room for domain-specific OCR improvements. This dataset enables robust benchmarking and reproducible research in ID verification, fraud detection, and Know Your Customer workflows, with tools to extend the dataset and task coverage as needed.

Abstract

Identity document (ID) image analysis has become essential for many online services, like bank account opening or insurance subscription. In recent years, much research has been conducted on subjects like document localization, text recognition and fraud detection, to achieve a level of accuracy reliable enough to automatize identity verification. However, there are only a few available datasets to benchmark ID analysis methods, mainly because of privacy restrictions, security requirements and legal reasons. In this paper, we present the DocXPand-25k dataset, which consists of 24,994 richly labeled IDs images, generated using custom-made vectorial templates representing nine fictitious ID designs, including four identity cards, two residence permits and three passports designs. These synthetic IDs feature artificially generated personal information (names, dates, identifiers, faces, barcodes, ...), and present a rich diversity in the visual layouts and textual contents. We collected about 5.8k diverse backgrounds coming from real-world photos, scans and screenshots of IDs to guarantee the variety of the backgrounds. The software we wrote to generate these images has been published (https://github.com/QuickSign/docxpand/) under the terms of the MIT license, and our dataset has been published (https://github.com/QuickSign/docxpand/releases/tag/v1.0.0) under the terms of the CC-BY-NC-SA 4.0 License.

DocXPand-25k: a large and diverse benchmark dataset for identity documents analysis

TL;DR

The paper tackles the lack of large, diverse public datasets for identity document analysis by introducing DocXPand-25k, a synthetic yet richly labeled collection of 24,994 IDs drawn from nine vector templates and placed into diverse real backgrounds. It provides comprehensive annotations for classification, localization, barcode/face detection, and OCR text fields, along with an open-source generation toolkit. Perceptual realism is validated via a LPIPS-based comparison to real-ID captures, and baseline experiments demonstrate practical utility with lightweight models for classification and localization, and a Tesseract-based OCR benchmark that remains far from industrial adequacy, indicating room for domain-specific OCR improvements. This dataset enables robust benchmarking and reproducible research in ID verification, fraud detection, and Know Your Customer workflows, with tools to extend the dataset and task coverage as needed.

Abstract

Identity document (ID) image analysis has become essential for many online services, like bank account opening or insurance subscription. In recent years, much research has been conducted on subjects like document localization, text recognition and fraud detection, to achieve a level of accuracy reliable enough to automatize identity verification. However, there are only a few available datasets to benchmark ID analysis methods, mainly because of privacy restrictions, security requirements and legal reasons. In this paper, we present the DocXPand-25k dataset, which consists of 24,994 richly labeled IDs images, generated using custom-made vectorial templates representing nine fictitious ID designs, including four identity cards, two residence permits and three passports designs. These synthetic IDs feature artificially generated personal information (names, dates, identifiers, faces, barcodes, ...), and present a rich diversity in the visual layouts and textual contents. We collected about 5.8k diverse backgrounds coming from real-world photos, scans and screenshots of IDs to guarantee the variety of the backgrounds. The software we wrote to generate these images has been published (https://github.com/QuickSign/docxpand/) under the terms of the MIT license, and our dataset has been published (https://github.com/QuickSign/docxpand/releases/tag/v1.0.0) under the terms of the CC-BY-NC-SA 4.0 License.
Paper Structure (19 sections, 5 equations, 8 figures, 5 tables)

This paper contains 19 sections, 5 equations, 8 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Related work
  3. Dataset
  4. Study of real documents
  5. Vectorial templates
  6. Document image generation
  7. Document image integration in backgrounds
  8. Dataset format and labels
  9. Statistics
  10. Visual similarity
  11. Metrics and baseline experiments
  12. ID classification
  13. ID localization
  14. Text fields recognition (OCR)
  15. Conclusion
  16. ...and 4 more sections

Figures (8)

  • Figure 1: Six sample documents of the proposed DocXPand-25k dataset.
  • Figure 2: Specimens of a few ID classes. Top line, left to right: French identity card (front and back sides), German passport (bio-data page). Bottom line, left to right: Spanish residence permit (front and back sides), Dutch passport (bio-data page).
  • Figure 3: Some templates designed for DocXPand.
  • Figure 4: Document image generation process.
  • Figure 5: Document image integration in backgrounds process.
  • ...and 3 more figures