Abstract
This study proposes a technique based on deep learning that utilizes complex neural network architectures to enhance the precision and reliability of optical character recognition (OCR) in barcode systems. The primary objective is to develop a dependable OCR (Optical Character Recognition) extractor system that can accurately identify alphanumeric characters in barcodes, even when those characters are distorted or partially obscured. The difficulty derives from the fact that conventional OCR algorithms aren’t up to the task of dealing with real-world barcode scanning challenges, such as fluctuating illumination, image noise and geometric distortions. We use Vision Transformers (ViT) and Convolutional Neural Networks (CNN) to extract features and classify characters to overcome these obstacles. As a result of its global attention method, which better collects contextual information, ViT obtained 97% accuracy, surpassing CNN’s 96% performance, which demonstrated good local feature recognition. To make sure the models can handle multiple formats and degrees of noise, they completed training and evaluation on an extensive dataset of barcode images. The results prove that ViT provides a more precise and extensible method for OCR in barcode systems. Finally, our study shows that OCR performance in retail and industrial settings may be greatly enhanced by deep learning, particularly with transformer-based models, where accuracy and speed are paramount.
Keywords: Barcode Recognition, Convolutional Neural Networks, Deep Learning, OCR Extraction, Optical Character Recognition, Vision Transformer.