Reviving the colors of paintings by removing the protective coating: a physical and virtual intervention

Abstract

When an existing coating over a painting is detrimental to its reading and full appreciation, it needs to be removed. Coating removal to reveal the underlying painting, with minimal physical intervention, may provide additional information regarding the painting and guidance for its restoration or intervention.

A Neural Network (NN) was devised to simulate the removal of a painting’s coating, using as training data a small area of the painting where the coating had been physically removed. Simulations of coating removal using the NN and two additional methodologies were compared to actual physical removal.

Hyperspectral images of the paintings with and without coating were acquired, and chromatic variations were computed by estimating differences in just-noticeable different colors (JND) values and in the color gamut, using CIECAM16-UCS. Comparisons were made between paintings with and without coating, and between paintings without coating and their simulations.

Results showed that removing the coating led to an increase in JND values (1.8 times on average) and in the color gamut, but the magnitude was dependent on the initial condition of the coating. When simulating coating removal, the NN produced the best chromatic simulation, with an average JND of approximately 2.6 ± 0.5 (1.1 ± 0.2 excluding lightness), while other methodologies produced differences of approximately 8.6 ± 5.7 (3.7 ± 3.0 excluding lightness).

Results achieved with the NN highlight its capability for simulating coating removal with minimal physical intervention to the painting, a valuable tool when complete coating removal without outcome prediction would be undesirable.