🏦 Enhanced Bank Check Security: Novel Dataset & Transformer-Based Signature Verification

📌 Accepted at the Document Analysis Systems (DAS) Workshop at ICDAR 2024
📄 Read the Paper
✍️ Muhammad Saif Ullah Khan*, Tahira Shehzadi*, Rabeya Noor, Didier Stricker & Muhammad Zeshan Afzal
( Equal contribution)

---

🔍 Overview

Automated signature verification on bank checks is crucial for fraud prevention and transaction security. However, real-world bank checks contain complex layouts with both textual and graphical elements, making this task particularly challenging.

This repository provides:
✅ Synthetic Signature Bankcheck Images (SSBI) Dataset – A realistic dataset with genuine & forged signatures on bank checks.
✅ Detection-Based Signature Verification Approach – Treats genuine & forged signatures as separate object detection classes.
✅ DINO-Based Transformer Model with a Dilation Module – Achieves 99.2% AP for genuine and 99.4% AP for forged signatures.
✅ Data Processing Scripts – Easily analyze, visualize, and expand the dataset.

---

📊 SSBI Dataset

The SSBI Dataset contains 4,360 annotated check images with genuine and forged signatures from 19 different signers. Each check also includes bounding boxes for:
✅ Signatures (both genuine & forged)
✅ Date field
✅ Amount fields (legal & courtesy)
✅ Payee field

📥 Download Dataset: SSBI v1.0.0
📌 Extract to: data/ssbi

🗂 Dataset Statistics

Table 1: Data Split for Genuine & Forged Signatures

Signature Type	Train	Validation	Total
✅ Genuine	2,352	1,008	3,360
❌ Forged	700	300	1,000
Total	3,052	1,308	4,360

Table 2: Object Size Distribution

Category	Train (Small/Medium/Large)	Val (Small/Medium/Large)
💰 Amount (Legal)	440 / 1,881 / 731	169 / 813 / 326
💵 Amount (Courtesy)	0 / 1,142 / 1,910	0 / 481 / 827
📅 Date	163 / 1,837 / 754	57 / 777 / 334
🏦 Payee	0 / 994 / 717	0 / 387 / 300
✍️ Signature (Forged)	2 / 304 / 394	0 / 122 / 178
✍️ Signature (Genuine)	4 / 915 / 1,433	1 / 358 / 649

🖼 Example Check Images:

---

📌 How to Use

1️⃣ Installation

Clone the repository and install dependencies:

git clone https://github.com/saifkhichi96/ssbi-dataset.git
cd ssbi-dataset
python3 -m venv venv
source venv/bin/activate
pip install -U pip wheel
pip install -r requirements.txt

2️⃣ Dataset Statistics

To analyze the dataset statistics, run:

python print_statistics.py

3️⃣ Dataset Structure

data/
│── ssbi/             # NOTE: Download & extract the dataset here
│   ├── annotations/  # COCO-style JSON annotations
│   ├── train/        # Training check images with both genuine & forged signatures
│   ├── val/          # Validation check images

🔄 Generating Synthetic Data

If the download link is unavailable or you wish to create more synthetic data, use:

python create_dataset.py

This script relies on the underlying data provided in the data/sources directory.

data/
|── sources/          # Underlying sources for the synthetic dataset
│   ├── checks/       # Original check images
│   ├── dummy_text/   # Dummy text for date, amount, and payee fields
│   ├── signatures/   # Original collection sheets with signature data
│   │   ├── genuine/  # Genuine signatures
│   │   ├── forged/   # Skilled forgeries

📜 Citation

If you use this dataset or method, please cite our paper:

@inproceedings{khan2024enhanced,
  title={Enhanced Bank Check Security: Introducing a Novel Dataset and Transformer-Based Approach for Detection and Verification},
  author={Khan, Muhammad Saif Ullah and Shehzadi, Tahira and Noor, Rabeya and Stricker, Didier and Afzal, Muhammad Zeshan},
  booktitle={International Workshop on Document Analysis Systems},
  pages={37--54},
  year={2024},
  organization={Springer}
}

📧 Contact

For questions, please reach out: 📩 Muhammad Saif Ullah Khan – saifkhichi96

License

This code and dataset in this repsitory are licensed under the CC BY-NC 4.0 license as described in the LICENSE file.

---

🔒 Enhancing Bank Check Security with Cutting-Edge AI! 🔍✨