DocHighlight: A Real-World Dataset for Document Specular Highlight Removal

📖 Introduction

DocHighlight is a large-scale, high-resolution dataset for document specular highlight removal, captured with a polarization-based acquisition pipeline across diverse real-world scenarios. The dataset is detailed in "Towards Real-World Document Specular Highlight Removal: The DocHighlight Dataset and DocSHRNet Method" (PRCV 2025), and the reference implementation DocSHRNet is available at 👉 https://github.com/shallweiwei/DocSHRNet.

---

📥 Download

The dataset is available via the following links:

• Baidu Netdisk
• Quark Netdisk

---

📌 Key Features

• 2,201 rigorously aligned highlight vs. highlight-free image pairs
• Average resolution of 2924 × 3672 (range: 1034×737 – 3468×4624)
• Covers books, magazines, multilingual text, and graphical content
• Captures real-world variations in document pose, illumination, and three camera devices
• Combines polarization imaging with manual quality verification for reliable ground truth

---

📝 Usage Notes

• 🔒 Non-commercial use only (CC BY-NC-SA 4.0).

---

📚 Citation

If you find this dataset useful in your research, please consider citing our paper.