TextBite: A Historical Czech Document Dataset for Logical Page Segmentation

2025-03-20

AI Summary

This work addresses logical page segmentation in historical Czech documents, proposing a purely image-based approach to eliminate OCR dependency and mitigate geometric distortion effects on evaluation. We introduce the first large-scale historical Czech logical layout dataset—spanning newspapers, dictionaries, and manuscripts from the 18th–20th centuries—comprising 8,449 pages and 78,863 paragraph-level annotations. We formally define the image-level logical segmentation task and propose a foreground-text-pixel-only evaluation paradigm. Methodologically, we decouple logical segmentation from OCR and geometric priors: integrating U-Net or Mask R-CNN for text detection, graph neural networks to model inter-paragraph semantic relationships, and handwriting-robust preprocessing. Our method achieves a paragraph-level F1 score of 0.82 on TextBite, outperforming OCR-dependent baselines by 17%. We publicly release the dataset, code, and evaluation framework, establishing the first dedicated benchmark for Central European historical document analysis.

Technology Category

• Natural Language Processing: Sentence-level Semantics, Textual Inference, etc.
• Computer Vision: Segmentation
• Machine Learning: Calibration & Uncertainty Quantification

Application Category

• Web Mining and Content Analysis: Large pretrained models with web data
• Graph Algorithms and Modeling for the Web: Algorithms and analysis for incomplete, noisy, or partially observed Web-related graphs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Logical page segmentation is an important step in document analysis, enabling better semantic representations, information retrieval, and text understanding. Previous approaches define logical segmentation either through text or geometric objects, relying on OCR or precise geometry. To avoid the need for OCR, we define the task purely as segmentation in the image domain. Furthermore, to ensure the evaluation remains unaffected by geometrical variations that do not impact text segmentation, we propose to use only foreground text pixels in the evaluation metric and disregard all background pixels. To support research in logical document segmentation, we introduce TextBite, a dataset of historical Czech documents spanning the 18th to 20th centuries, featuring diverse layouts from newspapers, dictionaries, and handwritten records. The dataset comprises 8,449 page images with 78,863 annotated segments of logically and thematically coherent text. We propose a set of baseline methods combining text region detection and relation prediction. The dataset, baselines and evaluation framework can be accessed at https://github.com/DCGM/textbite-dataset.

Problem

• Logical page segmentation without relying on OCR
• Evaluation metric using only foreground text pixels
• Dataset for historical Czech document segmentation

Innovation

• Image domain segmentation without OCR
• Foreground text pixel evaluation metric
• Baseline methods combining detection and prediction