Enhanced Bank Check Security: Introducing a Novel Dataset and Transformer-Based Approach for Detection and Verification

TL;DR

A novel dataset specifically designed for signature verification on bank checks is introduced, and a novel approach for writer-independent signature verification using an object detection network is proposed, effectively handling both detection and verification.

Abstract

Automated signature verification on bank checks is critical for fraud prevention and ensuring transaction authenticity. This task is challenging due to the coexistence of signatures with other textual and graphical elements on real-world documents. Verification systems must first detect the signature and then validate its authenticity, a dual challenge often overlooked by current datasets and methodologies focusing only on verification. To address this gap, we introduce a novel dataset specifically designed for signature verification on bank checks. This dataset includes a variety of signature styles embedded within typical check elements, providing a realistic testing ground for advanced detection methods. Moreover, we propose a novel approach for writer-independent signature verification using an object detection network. Our detection-based verification method treats genuine and forged signatures as distinct classes within an object detection framework, effectively handling both detection and verification. We employ a DINO-based network augmented with a dilation module to detect and verify signatures on check images simultaneously. Our approach achieves an AP of 99.2 for genuine and 99.4 for forged signatures, a significant improvement over the DINO baseline, which scored 93.1 and 89.3 for genuine and forged signatures, respectively. This improvement highlights our dilation module's effectiveness in reducing both false positives and negatives. Our results demonstrate substantial advancements in detection-based signature verification technology, offering enhanced security and efficiency in financial document processing.

Figures (5)

• Figure 1: Dataset Creation Pipeline. We use a semi-automated process for dataset generation comprising of a manual signature acquisition step--for obtaining forged and genuine signatures--and an automated step for inserting these signatures and filling other fields on bank checks.
• Figure 2: Comparative Analysis of Signature Samples - The two left columns display authentic signatures from an individual, showing the natural variations in their handwriting. The right column presents a forged signature sample, illustrating the outcome of a highly skilled forgery attempt with minimal visual differences from the genuine signatures. For each individual, eight forgeries were created: four executed with a ballpoint pen and four with a pencil to capture the diverse techniques used in forgery attempts.
• Figure 3: Our Bank Check Data Samples offer a diverse collection of check designs and ink colors, mimicking the variety banks handle daily. With everything from basic blue and green to intricate patterns, it challenges signature verification by altering signature visibility. The range of ink colors tests detection capabilities across different contrasts. Essential for creating algorithms that accurately detect forgeries, this dataset is key to enhancing transaction security.
• Figure 4: Illustration of bank checks before and after a dilation transformation
• Figure 5: Overview of our proposed approach for bank checks. The process begins with the input of two cashier's checks, each with a signature. The top check is fed into a Training Module, which consists of a DINO network dino23 with query selection, encoder, and decoder layers, and a hybrid matching component, where the model learns to predict the authenticity of the signature through supervised learning. After dilation transformation, the bottom check is processed through a guiding module, which parallels the Training Module's architecture but focuses on guiding the training process toward more stable and generalizable feature extraction.