Scheme for palimpsests reconstruction using synthesized dataset

Boraq Madi, Reem Alaasam, Raed Shammas, Jihad El-Sana

Research output: Contribution to journalArticlepeer-review


This paper presents Palimpsest Manuscripts Reconstruction Generative Adversarial Network, a novel framework for restoring the original forms of input palimpsests; it removes the over-text and refills the missing gaps in the text and background in an end-to-end manner. The structure and attributes of the under-text are encoded using reference patches. The generator network combines the encoded reference with an input palimpsest patch and restores the original form. To train our model, we synthesize palimpsests that mimic the attributes of the original ones. We compare the performance of our model with the state-of-art models using five different evaluation metrics, such as PSNR and SSIM. We show that our approach not only achieves state-of-the-art performance in terms of PSNR/SSIM metrics but also significantly improves the visual quality of the restored images.

Original languageEnglish
Pages (from-to)211-222
Number of pages12
JournalInternational Journal on Document Analysis and Recognition
Issue number3
StatePublished - 1 Sep 2023


  • Arabic documents
  • GANS
  • Hebrew documents
  • Occlusions
  • Palimpsest

ASJC Scopus subject areas

  • Software
  • Computer Vision and Pattern Recognition
  • Computer Science Applications


Dive into the research topics of 'Scheme for palimpsests reconstruction using synthesized dataset'. Together they form a unique fingerprint.

Cite this