Improving Handwritten Character Recognition Using YOLO Algorithms

Handwritten data recognition (HDR) remains a challenging task in real-world applications. The motivation for this work is automation of payment process from scanned payment slips to billing itself. This paper proposes an approach that leverages deep learning techniques to improve the accuracy and efficiency of HDR. Our methodology involves training a YOLO-based model on a dataset of real-world payment slips, annotated using the Computer Vision Annotation Tool (CVAT). The system is optimized for detecting and classifying handwritten digits without segmentation, making it more effective in handling overlapping and irregularly shaped digits. The proposed approach is implemented using Python, Ultralytics, and Torch. Experimental results demonstrate significant improvements in recognition accuracy compared to conventional OCR techniques. The model successfully detects handwritten digits with high confidence, making it a viable solution for automating financial document processing. This research contributes to advancing deep learning-based handwriting recognition and lays the foundation for further applications in document automation.