Wanglong Lu

Wanglong Lu

Nasdaq

DocPure: Prompt-Free Unified Document Restoration via Degradation-Aware Structure-Guided Wavelet Modulation

IEEE Transactions on Circuits and Systems for Video Technology, 2026 · Co-first author

Lingming Su, Wanglong Lu, Tao Wang, Kaihao Zhang, Nan Zhang, Liyan An, Hanli Zhao

DocPure framework

Brief description:

DocPure is a prompt-free unified framework for restoring document images affected by diverse degradations. It uses a degradation-aware structure auto-encoder to predict clean structural priors from degraded inputs, then combines spatial semantics with frequency-domain features through structure-guided wavelet interaction. This design enables one model to handle multiple restoration tasks without manual task prompts during inference.

Highlights:

  • Degradation-aware structure extraction through input-conditioned feature routing.
  • Structure-guided wavelet interaction between spatial and frequency-domain representations.
  • Cross-frequency adaptive modulation for recovering high-frequency text details while preserving document structure.
  • Unified restoration across deblurring, denoising, compression artifact reduction, and deshadowing.
  • Degradation labels are used as auxiliary training supervision rather than inference-time prompts.

[paper] [arxiv] [pdf] [html] [doi] [github]

Recommended citation:

@article{su2026docpure,
  title={DocPure: Prompt-Free Unified Document Restoration via Degradation-Aware Structure-Guided Wavelet Modulation},
  author={Su, Lingming and Lu, Wanglong and Wang, Tao and Zhang, Kaihao and Zhang, Nan and An, Liyan and Zhao, Hanli},
  journal={IEEE Transactions on Circuits and Systems for Video Technology},
  year={2026},
  doi={10.1109/TCSVT.2026.3720433}
}