Compact text recognizer improves historical text recognition with pretraining
A Free Lunch? Adapting PP-OCRv6 for Historical Text Recognition
Computer Vision and Pattern Recognition
Summary
Recognizing old handwritten texts is hard because big AI models are too costly and can make mistakes. The authors study a small, efficient text recognizer called PP-OCRv6 to see if it works well on historical writing. They find that while training from scratch isn’t always better, training the model on diverse data first helps it recognize old texts more accurately. This approach can even beat some large models designed for historical text recognition.
What this means in practice
- •For digital archivists: Use compact and pretrained text recognizers to improve automatic transcription of historical manuscripts with limited computing resources.
- •For software developers for historical text tools: Deploy lightweight OCR models adapted with diverse pretraining to enhance the accuracy of handwritten text recognition in multilingual heritage documents.
Authors
Benjamin Kiessling
Abstract
Despite impressive reported scores, large vision-language models have seen limited practical uptake in historical automatic text recognition because of their computational cost, dependence on large-scale pretraining, and hallucination. Historical ATR therefore continues to rely largely on compact CRNN line recognizers, which are visually grounded and trainable on modest data. Lightweight recurrence-free recognizers promise the accuracy of larger models with the practical advantages of CRNNs, yet have not been comprehensively evaluated on historical writing. We adapt PP-OCRv6, a recent compact text recognizer without strong language modeling, for historical line recognition and compare it with a conventional CRNN across generalized pretraining, domain-specific training, corpus-level fine-tuning, and manuscript-specific few-shot adaptation on multilingual Latin- and Arabic-script material. While PP-OCRv6 does not consistently outperform the baseline when trained from scratch, heterogeneous pretraining produces markedly better generalization. Comparisons with the Qwen3.5-based Medusa recognizer further show that fine-tuned PP-OCRv6 can outperform a large VLM tailored towards historical Latin-script HTR.