Text region detection in historical astronomical diagrams

2026-06-14 · Source: Computer Vision and Pattern Recognition · Field: Technology & Digital — Artificial Intelligence & Machine Learning · Depth: Expert, quick

Summary

A new large-scale, open-access dataset for text region detection in historical astronomical diagrams has been introduced, addressing a gap in existing benchmarks for mathematical diagrams. The dataset comprises 948 diagrams from the 8th to 18th centuries, featuring 10,940 oriented polygonal text regions across seven linguistic traditions including Arabic, Chinese, Byzantine, Latin, Hebrew, and Sanskrit. Each text instance is precisely annotated with ordered polygons encoding reading direction, and 2,293 Latin regions include 20 class labels. Researchers evaluated strong baselines like TESTR and DeepSolo++, alongside Poly-DETR, a DINO-DETR extension designed for ordered polygon vertex prediction. Poly-DETR achieved state-of-the-art performance on MTHv2 and cBAD2019 benchmarks, providing a solid baseline for this new dataset.

Key takeaway

For Computer Vision Engineers developing text detection solutions for historical documents, this new dataset and the Poly-DETR baseline offer a significant resource. You should consider integrating this open-access dataset into your training pipelines to improve model robustness for diverse historical diagram styles and linguistic traditions. Applying Poly-DETR's performance on oriented polygonal text regions can enhance accuracy in challenging historical document analysis tasks.

Key insights

A new large-scale dataset and Poly-DETR baseline advance text detection in historical astronomical diagrams.

Principles

• Specialized datasets are crucial for accurate text detection in niche historical document types.
• Precise oriented polygonal annotations are essential for complex text regions.

Method

Poly-DETR, an extension of DINO-DETR, predicts ordered polygon vertices to delineate text regions, achieving state-of-the-art performance on relevant benchmarks.

In practice

• Utilize the open-access dataset to train and benchmark models for historical document analysis.
• Implement Poly-DETR as a robust baseline for oriented text detection tasks.

Topics

• Text Detection
• Historical Documents
• Astronomical Diagrams
• Computer Vision
• Deep Learning
• Poly-DETR

Best for: AI Scientist, Computer Vision Engineer, Research Scientist

Related on AIssential

• How to Accurately Extract Structured Data from Complex Documents Using AI — Combining document parsing and structured extraction into a single LLM call improves accuracy for complex layouts.
• FusionRS: A Large-Scale RGB-Infrared Remote Sensing Dataset for Dual-Modal Vision-Language Foundation Models — FusionRS enables dual-modal RGB-infrared vision-language models through a novel dataset with IR-aware captions.
• DETR-ViP: Detection Transformer with Robust Discriminative Visual Prompts — DETR-ViP enhances visual prompted object detection by improving prompt discriminability and fusion.
• [R] Can Vision-Language Models See Squares? Text-Recognition Mediates Spatial Reasoning Across Three Model Families — VLMs excel at text-based spatial tasks but struggle with non-textual visual patterns, indicating a strong implicit OCR pipeline.
• SADGE: Structure and Appearance Domain Gap Estimation of Synthetic and Real Data — Synthetic data utility for computer vision tasks is best predicted by the non-linear interplay of appearance and geometric similarities, not either alone.
• Archivists Turn to LLMs to Decipher Handwriting at Scale — General-purpose LLMs now reliably transcribe historical handwriting, outperforming specialized software in speed, cost, and accuracy.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Computer Vision and Pattern Recognition.