SteveBronder / platypus

remove cradling artifacts in X-ray images of paintings on panel

IPOL Cradle Detection and Removal code package

This program implements an algorithmic solution for detecting and correcting the visually displeasing effects of cradling in X-ray images of panel paintings. The code is supplied in addition to the IPOL documentation material, available online at http://www.ipol.im.

== Authors ==

The program is written by Gabor Fodor [e-mail], based heavily on the Matlab code of Rujie Yin [e-mail] and with guidance from Bruno Cornelis [e-mail] and Ingrid Daubechies [e-mail].

== License ==

Distributed under the terms of the BSD license. See the file license.txt for details.

== Contents ==

img sample images folder src source code folder license.txt license file (GPL v3+)

== Requirements ==

This code needs OpenCV 4+ in order to compile. Additionally, compiling is recommended with gcc 4.7.2+, older compiler versions were not tested.

To obtain OpenCV, please refer to http://opencv.org/ OpenCV is released under a BSD license and hence it’s free for both academic and commercial use.

== Compilation ==

Make the executables with the following

cmake -S . -B "build" -DCMAKE_BUILD_TYPE=RELEASE
cd build
make -j2 all

The executables Demo, CradleRemove, and TextureRemoval will be available in the build/bin directory.

== Usage ==

Three executable are built, mainCradleRemoval and mainTextureRemoval, demonstrating the two functionalities of the code in separate executable, and mainDemo that performs full scale cradle and wood-grain separation across the entire X-ray.

mainCradleRemoval

This piece of demo code demonstrates cradle removal for an image specified by the user. Given an X-ray image of a painting containing the cradle artefact, the code attempts to remove it from the image. It first detects the approximate position of the cradle pieces and then fits a multiplicative correction model that attempts to rectify the pixel intensity of cradled pieces. NOTE: we do not check if the image specified does not exist, and the code will result in an error.

Parameters:

• path+filename of input image to be processed
• path+filename of output images (optional; defaults to "out" if not specified)
• and number of horizontal and vertical pieces in the cradled image (optional)

Output:

• _original.png original, unprocessed image file saved as grayscale png
• _nointensity.png x-ray with cradle removed, saved as grayscale png
• _cradle.png cradle component of the separation, saved as grayscale png
• _mask.png cradle mask file, marking location of cradle pieces
• .msf an auxiliary file containing processing information, usable for texture removal

The code handles RGB input (by converting it to grayscale), but all output will be grayscale.

mainTextureRemoval

This piece of demo code demonstrates a refining step in the cradle removal process where the wood-grain component of the cradle is targeted. The intensity adjustment still leaves texture parts of the cradle in the processed image, namely, wood-grain structures that are not part of the painting or the supporting wooden panel. Using a Bayesian supervised learning technique, a statistical model is built that attempts to separate the wood-grain component from the X-ray image.

NOTE: this code requires an already processed X-ray image file, as the one provided by mainCradleRemoval and the *.msf file that contains all processing information. This demo code was set up to run in combination with mainCradleRemoval, but upon tweaking the code one can separate the two functionalities from one another.

Parameters:

• path+filename of input image to be processed
• X coordinate of the 464x464 block where wood-grain separation should be executed
• Y coordinate of the 464x464 block where wood-grain separation should be executed

Output:

• "inc.png" a 464×464 crop of the original
• "out.png" result after cradle removal, without wood-grain separation
• "out2.png" result after cradle removal and wood-grain separation
• "cradle.png" cradle component, without wood-grain separation
• "cradle2.png" cradle component, after wood-grain separation

mainDemo

This piece of demo code demonstrates the combined cradle removal and wood-grain separation for an image specified by the user. If the image specified does not exist, the code results in an error.

Parameters:

• arg[1] path+filename of input image to be processed

Output:

• "solution.png" x-ray with cradle and wood-grain removed, saved as grayscale png
• "difference.png" cradle component, before wood-grain removal
• "difference_textrem.png" cradle component, after wood-grain removal

== Use examples ==

For a full-scale cradle and wood-grain separation, run mainDemo with the input image as argument Example:

./mainCradleRemoval /img/Ghissi.png

Note that this is could take very long, so for a fast execution of the demo code run files mainCradleRemoval and mainTextureRemoval one after the other. Example:

#with specifying number of cradle pieces ./mainCradleRemoval /img/Ghissi.png outfile 1 1

#texture separation in a 464x464 block centered at position (200,200) ./mainTextureRemoval outfile 200 200

OR

#without specifying number of cradle pieces ./mainCradleRemoval /img/stjerome_detail1.png outfile2 #texture separation in a 464x464 block centered at position (200,200) ./mainTextureRemoval outfile2 400 500