, ,


A painting degrades overtime because of different chemical and physical processes.  The degradation may be due to some dirt or dust, or light exposure which hastens the usual deterioration of the chemicals used in the paint.

Physical cleaning of the painting is one of the ways to bring the painting back to its original state.  A problem though arises because physical cleaning of the painting is mostly subjective. There are no clear standards of restoration and it is pratically trial and error.  This is particularly risky if  the painting is a treasured heritage such as an Amorsolo.

Malacañan by the River by Fernando Amorsolo from the UP Vargas Museum Collection (oil on canvas, 1948) before (above) and after (below) after digital cleaning and context-based post processing. Courtesy: CMT Palomero and MN Soriano, Opt. Express 19 (2011).

This is where digital cleaning becomes important. Digital cleaning tries to find a mathematical form that describes the dirtying process.  The painting can then be digitally restored by reversing this process.  There are several ways to do this but mostly these techniques need a little invasive procedure such as physically cleaning a small patch or getting a small sample to get fresh paints.

Cherry May T. Palomero and Maricor N. Soriano offer a way to clean a painting through a new digital cleaning technique in their recent publication in Optics Express [1]. By looking at unexposed portions of the painting, they are able to restore the Malacañan by the River of Amorsolo to its ‘original’ colors without an invasive procedure.

The authors used neural network “to learn the transformation of pixel values from dirty to clean.”  The training data they used is the part of the painting that has both an unexposed and exposed portion. The neural network implementation will not be discussed here in detail but the authors used a “standard two-layered feed forward neural network trained using a Levenberg-Marquardt optimization.”  In using neural network, the authors was able to forgo the invasive procedure inherent in other digital cleaning technique.

A problem they encounter in their cleaning process is ‘overcleaning’.  This manifests as paint patches appearing to have flaked off.

They solved this problem with a context-based processing.  They did this by looking at the color distribution before and after the cleaning process.  Beyond a defined certain threshold, the new color distribution is reverted back to the original distribution. This solves the overcleaning.

So how does the dirt of the painting look like?

Visualization of the dirt layer. Courtesy: CMT Palomero and MN Soriano, Opt. Express 19 (2011).

The dirt layer was visualized by taking the spectral information before and after the cleaning process. This results to the dirt spectra seen on the right. The dirt spectra agrees very well with the yellow oxidized varnish and the brown dirt and grime that the painting had accumulated over time.  The dirt spectra seen is also not homogenous or even althrough out the painting.

As one can see, the painting of Amorsolo looks more vibrant when it is cleaned enabling viewers to enjoy it in its original splendor.


[1] Palomero, C., & Soriano, M. (2011). Digital cleaning and “dirt” layer visualization of an oil painting Optics Express, 19 (21) DOI: 10.1364/OE.19.021011 (Open Access!)

[2] The work is sponsored by the University of the Philippines Open Grant project no. 062929OG.