Authors of this software: Simone Parisotto and Luca Calatroni
Other authors and collaborators: Carola-Bibiane Schönlieb, Stella Panayotova, Paola Ricciardi
Version 1.0
Date: 22/02/2018
This is a companion software for the journal article:
L. Calatroni, M. D’Autume, R. Hocking, S. Panayotova, S. Parisotto, P. Ricciardi, C.-B. Schönlieb,
"Unveiling the invisible - mathematical methods for restoring and interpreting illuminated manuscripts"
Heritage Science, Springer International Publishing (2018) 6: 56. DOI: 10.1186/s40494-018-0216-z
Please use the following entry to cite the article:
@Article{CalAutHocPanParRicSch2018,
author="Calatroni, Luca and d'Autume, Marie and Hocking, Rob
and Panayotova, Stella and Parisotto, Simone
and Ricciardi, Paola and Sch{\"o}nlieb, Carola-Bibiane",
title="Unveiling the invisible: mathematical methods for restoring and interpreting illuminated manuscripts",
journal="Heritage Science",
year="2018",
volume="6",
number="1",
pages="56",
publisher="Springer International Publishing",
doi="10.1186/s40494-018-0216-z",
url="https://doi.org/10.1186/s40494-018-0216-z"
}
The workflow consists in two steps: the recognition of damaged areas in manuscripts and the inpainting restoration.
- 1st step: run Matlab file for segmentation combining active contour + kmeans
./manuscript_segmentation.m
- 2nd step: run bash script for the inpaint the damaged areas detected in step 1
./manuscript_inpainting.sh
The experiment with number XXX is stored in the folder “./results/paper_results/testXXX” folder, where
- 101,102,... refers to images from Manuscript 1;
- 201,202,... refers to images from Manuscript 2.
In folder ./results/paper_resultS/test101 we store the results of experiment 01 for Manuscript 1:
- input_orig101.png: the real crop of Manuscript 1;
- input101.png: the preprocessing (smoothing/texture removal);
- overlap_SUPER101.png: overlap between image crop and supervised pixel inputs in blue squares (just one pixel, dilated for display purposes);
- overlap_SUPERwithCV101.png: overlap between active contour regions in yellow and supervised pixel inputs in blue squares (just one pixel, dilated for display purposes);
- overlap_CV101png: overlap between image crop and active contour regions in yellow;
- overlap101.png: final overlap between image and segmentation from kmeans+activecontour
- mask101.png: final segmentation mask
- masked101.png: input for TV inpainting
- TVinpainted101.png: TV inpainting result
- PATCHinpainted101_PxP: results of nonlocal exemplar-based inpainting result with patch of size P and TV inpainting as initialization
The content and images of this work are part of the ILLUMINATED: Manuscripts in the making project. We refers to the project website for more information and the related copyright notice.