TEXT EXTRACTION FROM SUPERMARKET PRICE TAGS IN ENGLISH AND CYRILLIC WITH OPEN-SOURCE OCR

DOI: 10.31673/2412-4338.2025.048901

Abstract

Supermarket price tags often contain small, densely formatted text presented under challenging visual conditions such as glare, low contrast, irregular lighting, and the presence of both Latin and Cyrillic scripts. These factors significantly complicate text extraction, particularly for individuals with low vision, and reduce the effectiveness of general-purpose OCR systems commonly used in assistive applications. Reliable recognition of this text is crucial for tools intended to support independent navigation and price comprehension in retail environments. This article investigates the feasibility of using lightweight, open-source OCR engines to extract accurate and readable text from tightly cropped images of supermarket price tags in two scripts: English (Latin) and Cyrillic.

This study evaluates three publicly available OCR frameworks—Tesseract, EasyOCR, and PaddleOCR—selected for their widespread accessibility, broad multilingual support, and compatibility with resource-constrained devices such as smartphones and embedded systems. To ensure a representative and diverse benchmark, a unified dataset was compiled from multiple open-source collections and organized into Latin-only, Cyrillic-only, and mixed-language subsets. Each OCR model was tested under both clean conditions and a variety of synthetic distortions designed to mimic real-world retail environments, including blur, contrast degradation, and perspective warp.

The article provides a detailed description of the dataset construction process, preprocessing techniques used to enhance image quality, evaluation methodology, and the metrics applied to measure recognition reliability. It also addresses practical challenges encountered during the study, such as inconsistent or noisy annotations and the difficulties posed by mixed-script content frequently found in bilingual retail contexts. By examining the strengths, limitations, and robustness of each OCR engine, this work offers guidance for developers and researchers creating OCR-based assistive technologies aimed at improving retail accessibility for users with visual impairments.

Keywords: optical character recognition (OCR), price tag recognition, multilingual text, open-source models, text extraction, image analysis.