A Hidden Portrait by Edgar Degas

The preservation and understanding of cultural heritage depends increasingly on in-depth chemical studies. Rapid technological advances are forging connections between scientists and arts communities, enabling revolutionary new techniques for non-invasive technical study of culturally significant, highly prized artworks. We have applied a non-invasive, rapid, high definition X-ray fluorescence (XRF) elemental mapping technique to a French Impressionist painting using a synchrotron radiation source, and show how this technology can advance scholarly art interpretation and preservation. We have obtained detailed technical understanding of a painting which could not be resolved by conventional techniques. Here we show 31.6 megapixel scanning XRF derived elemental maps and report a novel image processing methodology utilising these maps to produce a false colour representation of a “hidden” portrait by Edgar Degas. This work provides a cohesive methodology for both imaging and understanding the chemical composition of artworks, and enables scholarly understandings of cultural heritage, many of which have eluded conventional technologies. We anticipate that the outcome from this work will encourage the reassessment of some of the world’s great art treasures.

| Portrait of a Woman mounted at the XFM beamline. The Maia 384A energy dispersive detector array is positioned in front of the painting with the Kirkpatrick-Baez (KB) mirror focusing optics located upstream. (Edgar Degas, French, 1834-1917, Portrait of a Woman (Portrait de femme) c. 1876-80, oil on canvas, 46.3 × 38.2 cm, National Gallery of Victoria, Melbourne, Felton Bequest, 1937).

Maia detector
False colour image reconstruction. Colour renderings improve the ability to observe and highlight features of interest. The false colour image reconstruction method makes it possible to exaggerate the intensity and highlight location of particular elements, or apply an entirely artificial colour to an element. Thus in the case of Portrait of a Woman it is possible to re-process the image to accentuate features from the lower or upper image, or a combination of both. It is possible to use this as a tool to aid the attribution of pigment identity. Where an element that is capable of producing a variety of coloured pigments, separate reconstructions can be made with the colour possibilities, and the resulting reconstruction compared to other works by the artist.
Processing of the data using commercially available software did not yield effective results even when only three elements were overlayed, generally giving mixed shades of brown and poorly defined reconstructed features. The authors are not aware of other systems which enable successful and semiautomated multi-element overlays, while retaining data integrity for large data sets.
For an elemental map, , the corresponding layer opacity, , is given by = ( 2.2 ) /2.2 , where is an transparency level and is a gamma correction value that acts like a contrast setting. These parameters help compress the high dynamic range measurements in the elemental maps down to the range visible on computer screens. Element layers were assigned colour by applying RGB values consistent with the proposed pigment identity. Transparency of the element layer was assigned to achieve a plausible reconstruction, such as, skin tones consistent with other Degas paintings from the period. Elements were chosen to be layers in the colour reconstruction if they were above trace levels, that is, they were likely to add to the visual reconstruction. The colour reconstructed image was not observed to impact if additional trace elements were included as additional layers. The ordering of the layers has an effect on the resulting image. The ordering was chosen subjectively in that it produced an image of similar appearance to other Degas works. For example, applying a high transparency layer over low for interpreting the artist's technique. The selection of a "wrong" colour for a layer (by proposing other pigment identities) gave a poor reconstruction; for example if the sitter had green hair, which was inconsistent with the artist's known practices. Different RGB combinations produce similar results, up to a point. The results are not particularly sensitive to the absolute choice of RGB values, for example, slightly adjusting the RGB values for the blue layer does not result in a significantly different reconstruction.
From the variety of calculated overlay reconstructions we selected the version which most closely matched how an unobscured artwork may have appeared. A limitation of this layering method is that each element can exist at only one location in the layer stack, theoretically reducing the quality of the final reconstruction (real paintings are not defined layers, but mixtures). We would hope to further refine our computational technique.

Figure S2 | Assignment of colour to elemental distributions.
Documented pigment colours can be assigned to individual elemental distributions. The software overlays the coloured maps to generate a composite false colour image situated on a white base layer . By altering the colour assigned to an elemental distribution is possible to accentuate features, and assess the plausibility of the assigned pigment. For example if Cr was designated orange (PbCrO 4 PbO) instead of green (Cr 2 O 3 ) the composite image would not resemble other works by the artist at period, and the pigment identity assignment is probably incorrect. The elements, colour assignment and opacity assigned to the elemental maps to produce the composite image published here are given in Table S1. (Edgar Degas, French, 1834-1917