Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Synthetic Craquelure Generation for Unsupervised Painting Restoration

Jana Cuch-Guillén, Antonio Agudo, Raül Pérez-Gonzalo

TL;DR

This work proposes a fully annotation-free framework driven by a domain-specific synthetic craquelure generator, which simulates realistic branching and tapered fissure geometry using B\'ezier trajectories and employs a detector-guided strategy, injecting the morphological map as an input spatial prior.

Abstract

Cultural heritage preservation increasingly demands non-invasive digital methods for painting restoration, yet identifying and restoring fine craquelure patterns from complex brushstrokes remains challenging due to scarce pixel-level annotations. We propose a fully annotation-free framework driven by a domain-specific synthetic craquelure generator, which simulates realistic branching and tapered fissure geometry using Bézier trajectories. Our approach couples a classical morphological detector with a learning-based refinement module: a SegFormer backbone adapted via Low-Rank Adaptation (LoRA). Uniquely, we employ a detector-guided strategy, injecting the morphological map as an input spatial prior, while a masked hybrid loss and logit adjustment constrain the training to focus specifically on refining candidate crack regions. The refined masks subsequently guide an Anisotropic Diffusion inpainting stage to reconstruct missing content. Experimental results demonstrate that our pipeline significantly outperforms state-of-the-art photographic restoration models in zero-shot settings, while faithfully preserving the original paint brushwork.

Synthetic Craquelure Generation for Unsupervised Painting Restoration

TL;DR

This work proposes a fully annotation-free framework driven by a domain-specific synthetic craquelure generator, which simulates realistic branching and tapered fissure geometry using B\'ezier trajectories and employs a detector-guided strategy, injecting the morphological map as an input spatial prior.

Abstract

Cultural heritage preservation increasingly demands non-invasive digital methods for painting restoration, yet identifying and restoring fine craquelure patterns from complex brushstrokes remains challenging due to scarce pixel-level annotations. We propose a fully annotation-free framework driven by a domain-specific synthetic craquelure generator, which simulates realistic branching and tapered fissure geometry using Bézier trajectories. Our approach couples a classical morphological detector with a learning-based refinement module: a SegFormer backbone adapted via Low-Rank Adaptation (LoRA). Uniquely, we employ a detector-guided strategy, injecting the morphological map as an input spatial prior, while a masked hybrid loss and logit adjustment constrain the training to focus specifically on refining candidate crack regions. The refined masks subsequently guide an Anisotropic Diffusion inpainting stage to reconstruct missing content. Experimental results demonstrate that our pipeline significantly outperforms state-of-the-art photographic restoration models in zero-shot settings, while faithfully preserving the original paint brushwork.
Paper Structure (14 sections, 7 equations, 7 figures, 3 tables)

This paper contains 14 sections, 7 equations, 7 figures, 3 tables.

Table of Contents

  1. INTRODUCTION
  2. RELATED WORK
  3. UNSUPERVISED CRAQUELURE ART PAINTING RESTORATION
  4. Crack Detection
  5. Learning-Based Brushstroke–Crack Segmentation
  6. Crack-Filling Methods
  7. SYNTHETIC CRAQUELURE DATA
  8. RESULTS
  9. Crack Detection
  10. Crack-Filling Methods
  11. Traditional Baseline Comparison
  12. Ablation Study of the Segmentation Modules
  13. Quantitative Comparison
  14. Conclusion

Figures (7)

  • Figure 1: Overview of the proposed unsupervised restoration algorithm. Cracks are first identified using morphological filtering, then refined using a model fine-tuned on synthetic data, and finally restored via inpainting.
  • Figure 2: Resulting top-hat mask without noise filtering (center) and with noise filtering (right).
  • Figure 3: Example of synthetic generated data: pairs of original paintings (left) and resulting damaged images (right).
  • Figure 4: Paintings with real cracks constituting our test set.
  • Figure 5: Global entropy comparison using the top-hat transform, evaluating Black Crack Removal (BCR) versus combined Black and White Crack Removal (B+WCR).
  • ...and 2 more figures