SD7K

High-resolution Document Shadow Removal Dataset and Network

SD7K is a comprehensive solution for document shadow removal, consisting of a large-scale real-world dataset and a frequency-aware shadow erasing network. This project addresses the practical problem of shadow removal in document digitization workflows.

Dataset Highlights

  • Large-scale: Over 7,000 high-resolution document images
  • Real-world diversity: Captured under various lighting conditions and document types
  • High quality: Carefully curated and annotated shadow regions
  • Practical relevance: Designed for real document processing applications

Technical Innovation

Frequency-Aware Shadow Erasing Network

Our network introduces several key innovations:

  1. Frequency Domain Analysis: Analyzes shadows in the frequency domain for better understanding
  2. Structure Preservation: Maintains document structure while removing shadows
  3. Multi-scale Processing: Handles shadows at different scales and intensities

Architecture Details

  • Encoder-Decoder Structure: Efficient feature extraction and reconstruction
  • Attention Mechanisms: Focus on shadow regions while preserving text quality
  • Loss Functions: Specialized loss functions for document-specific requirements

Applications

This work has practical implications for:

  • Document digitization workflows
  • Optical Character Recognition (OCR) systems
  • Archive preservation projects
  • Mobile document scanning applications

Performance

SD7K achieves state-of-the-art performance on document shadow removal benchmarks, significantly outperforming existing methods in both quantitative metrics and visual quality.

Publication

Published at ICCV 2023, one of the premier conferences in computer vision.

Citation:

Li, Z., Ke, Q., Bennamoun, M., & Boussaid, F. (2023). 
High-resolution Document Shadow Removal via A Large-scale Real-world Dataset 
and A Frequency-aware Shadow Erasing Net. 
IEEE/CVF International Conference on Computer Vision.