Beyond Simple Character Recognition: A Comparative Study of Vision-Language Models and Dedicated OCR Systems in Edge Cases

The rapid rise of Vision-Language Models (VLMs) has reframed OCR as a generative task, yet deployment risk is not captured by capped CER/WER alone. This study compares nine OCR systems across three paradigms. Seven full-suite systems—Tesseract, PaddleOCR, EasyOCR, DeepSeek-OCR-2, Chandra, Qwe...

Descrición completa

Detalles Bibliográficos

Principais autores:	Juran Kim, Chang-Young Lee, Chang-Yup Han, Nam-Hyun Yoo, Jinhong Yang
Formato:	Artigo
Idioma:	Inglês
Publicado:	IEEE 2026-01-01
Series:	IEEE Access
Assuntos:	Benchmark comparison edge case error taxonomy failure mode analysis hallucination optical character recognition (OCR)
Acceso en liña:	https://ieeexplore.ieee.org/document/11571805/
Tags:	Engadir etiqueta Sen Etiquetas, Sexa o primeiro en etiquetar este rexistro!