Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A Benchmark of State-Space Models vs. Transformers and BiLSTM-based Models for Historical Newspaper OCR

Merveilles Agbeti-messan, Thierry Paquet, Clément Chatelain, Pierrick Tranouez, Stéphane Nicolas

Abstract

End-to-end OCR for historical newspapers remains challenging, as models must handle long text sequences, degraded print quality, and complex layouts. While Transformer-based recognizers dominate current research, their quadratic complexity limits efficient paragraph-level transcription and large-scale deployment. We investigate linear-time State-Space Models (SSMs), specifically Mamba, as a scalable alternative to Transformer-based sequence modeling for OCR. We present to our knowledge, the first OCR architecture based on SSMs, combining a CNN visual encoder with bi-directional and autoregressive Mamba sequence modeling, and conduct a large-scale benchmark comparing SSMs with Transformer- and BiLSTM-based recognizers. Multiple decoding strategies (CTC, autoregressive, and non-autoregressive) are evaluated under identical training conditions alongside strong neural baselines (VAN, DAN, DANIEL) and widely used off-the-shelf OCR engines (PERO-OCR, Tesseract OCR, TrOCR, Gemini). Experiments on historical newspapers from the Bibliothèque nationale du Luxembourg, with newly released >99% verified gold-standard annotations, and cross-dataset tests on Fraktur and Antiqua lines, show that all neural models achieve low error rates (~2% CER), making computational efficiency the main differentiator. Mamba-based models maintain competitive accuracy while halving inference time and exhibiting superior memory scaling (1.26x vs 2.30x growth at 1000 chars), reaching 6.07% CER at the severely degraded paragraph level compared to 5.24% for DAN, while remaining 2.05x faster. We release code, trained models, and standardized evaluation protocols to enable reproducible research and guide practitioners in large-scale cultural heritage OCR.

A Benchmark of State-Space Models vs. Transformers and BiLSTM-based Models for Historical Newspaper OCR

Abstract

End-to-end OCR for historical newspapers remains challenging, as models must handle long text sequences, degraded print quality, and complex layouts. While Transformer-based recognizers dominate current research, their quadratic complexity limits efficient paragraph-level transcription and large-scale deployment. We investigate linear-time State-Space Models (SSMs), specifically Mamba, as a scalable alternative to Transformer-based sequence modeling for OCR. We present to our knowledge, the first OCR architecture based on SSMs, combining a CNN visual encoder with bi-directional and autoregressive Mamba sequence modeling, and conduct a large-scale benchmark comparing SSMs with Transformer- and BiLSTM-based recognizers. Multiple decoding strategies (CTC, autoregressive, and non-autoregressive) are evaluated under identical training conditions alongside strong neural baselines (VAN, DAN, DANIEL) and widely used off-the-shelf OCR engines (PERO-OCR, Tesseract OCR, TrOCR, Gemini). Experiments on historical newspapers from the Bibliothèque nationale du Luxembourg, with newly released >99% verified gold-standard annotations, and cross-dataset tests on Fraktur and Antiqua lines, show that all neural models achieve low error rates (~2% CER), making computational efficiency the main differentiator. Mamba-based models maintain competitive accuracy while halving inference time and exhibiting superior memory scaling (1.26x vs 2.30x growth at 1000 chars), reaching 6.07% CER at the severely degraded paragraph level compared to 5.24% for DAN, while remaining 2.05x faster. We release code, trained models, and standardized evaluation protocols to enable reproducible research and guide practitioners in large-scale cultural heritage OCR.

Paper Structure

This paper contains 26 sections, 8 equations, 3 figures, 7 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Historical Document OCR and Newspaper Challenges
  4. Sequence Modeling: CTC, Attention, and Transformers
  5. Paragraph-Level Recognition
  6. State-Space Models as Attention Alternatives
  7. Mamba-based OCR
  8. Preliminaries
  9. Mamba Formulation
  10. Mamba-based Architectures
  11. Shared CNN Encoder
  12. Bidirectional Mamba Context Modeling
  13. Decoding Variants
  14. Experimental Setup
  15. Datasets
  16. ...and 11 more sections

Figures (3)

  • Figure 1: Mamba-based OCR architecture.Shared visual encoder (blue) extracts features, bidirectional Mamba connector (purple) models global context, and three decoder variants (orange) implement different decoding strategies: autoregressive (AR), non-autoregressive with static queries (NAR), and CTC alignment.
  • Figure 2: Accuracy vs latency trade-off on line-level recognition. Left: Antiqua test set, Mamba-AR achieves DAN-level accuracy with 2.9× lower latency. Right: Fraktur test set, VAN and Mamba-NAR provide optimal speed-accuracy balance for Gothic script.
  • Figure 3: Peak inference memory vs sequence length. Mamba-AR scales linearly ($O(n)$) while DAN quadratically ($O(n^2)$), confirming theoretical predictions.