Authors: Marella Maheswari, P ASHOKA
Abstract: Revolutionizing the verification of bank checks, this innovative technology simplifies the process by integrating deep learning, image processing, and an intuitive Django-based web interface. It streamlines the process with little human participation, making it easier than ever before. Our Convolutional neural network (CNN) trained on the IDRBT check dataset and executed in PyTorch has a 99.14% success rate in recognizing handwritten digits, as shown in the introductory article. Adaptive thresholding and Gaussian blurring are implemented in the source code to enhance the picture preparation. The optical character recognition (OCR) in MATLAB can recover machine-printed text with 97.7 percent accuracy, including IFSC codes and account numbers, when Pytesseract is used in the code for region-based text extraction. The approach uses SVM classification and SIFT feature extraction for real-time authenticity checks, allowing signature verification powered by SIFT and SVM to reach 98.1% accuracy. The web-based interface allows more users to upload photos of checks, train models, see datasets, and get immediate categorization results ("Genuine" or "Not Genuine"). The system complies with CTS-2010 standards for Indian banks and the extraction of critical details such as signatures, amounts, and check numbers is possible even if it supports formats from other countries. In order to automate the verification process and decrease processing time, operational expenditures, and fraud risks, it makes use of contour detection and region-based analysis. This scalable solution sets a new standard for secure, efficient financial transactions by combining the rigors approach from the paper with the actual code implementation. Future versions may support more than one language and format.
DOI: http://doi.org/10.5281/zenodo.20286653
