PERFORMANCE EVALUATION OF TESSERACT, EASYOCR, AND TROCR MODELS FOR OPTICAL CHARACTER RECOGNITION SYSTEMS
Loading...
Date
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
UNIVERSITI MALAYSIA SARAWAK
Abstract
Description
This research evaluates the performance of three Optical Character Recognition (OCR) methods of Tesseract, EasyOCR, and TrOCR across the Chars74k and Total Text datasets. Through K-fold cross-validation, the study analyzes character and word error rates, inference times, and generalization capabilities. Results highlight the trade-off between traditional and deep learning approaches, with TrOCR excelling in challenging scene text, EasyOCR offering a balance between accuracy and efficiency, and Tesseract excelling on cleaner text. A public survey further explores perceptions of OCR’s usefulness and future relevance. Findings guide future improvements and practical deployments of OCR technologies.