Multi-stage deep learning framework for robust recognition of overlapping and faded handwritten text in bank cheques

Hitesh Chaitanyaswami ¹, Ashwin Dobariya ¹, Sailesh Iyer ²

⁰Faculty of Computer Applications, Marwadi University, Rajkot, Gujarat, India.

Scientific Reports +

|

December 3, 2025

View abstract on PubMed

Summary

This study introduces a deep learning framework for automated bank cheque analysis. It accurately extracts handwritten data and detects forged cheques, improving financial transaction efficiency.

Area Of Science

Computer Science
Artificial Intelligence
Image Processing

Background

Automated cheque processing is vital for financial institutions.
Existing systems struggle with overlapping text and degraded handwriting.
Manual processing is error-prone and inefficient.

Purpose Of The Study

To develop a robust deep learning framework for automated cheque field extraction.
To address challenges in differentiating overlapping text and restoring faded handwriting.
To accurately classify cheques as genuine or forged.

Main Methods

A multi-stage deep learning framework involving preprocessing, segmentation, and extraction.
Hybrid approach for segmenting overlapping handwritten and printed text.
Nanonet model for key field extraction and a CNN for forgery detection.

Main Results

Achieved high accuracy in segmenting overlapping text.
Successfully extracted key handwritten fields like date, name, and amount.
Demonstrated superior performance in recognition robustness and classification accuracy (98.79% for forgery detection).

Conclusions

The proposed framework effectively handles complex cheque image issues.
It significantly improves automated cheque processing accuracy and efficiency.
The method shows promise for real-world financial applications.

Keywords:

Forged cheques Genuine cheques Handwritten texts Sobel sparse Marr–Hildreth edge detector