Spectral-divergence based pigment discrimination and mapping: A case study on The Scream (1893) by Edvard Munch

Abstract

An important application of imaging spectroscopy or hyperspectral imaging is the classification or discrimination of pigments based on the obtained spectral reflectance information. As opposed to being a point-analysis tool, this non-invasive method captures the entire surface of interest. This means that its potential is not only in the discrimination of pigments but also in their mapping. However, the challenge lies in the fact that in a real painting, there is no clear-cut edge between regions with certain pure pigments or of the exact same mixture. Pigments and other paint materials mix seamlessly and continuously in the physical domain. In this article, we introduce a divergence-based approach to pigment discrimination and mapping. The methodology is then applied to Munch's masterpiece The Scream (1893), whose pigments and materials have been identified for several points in the painting in a previous study. Through the introduced methodology, we have been able to extend the point analyzes of pigments and materials to the entire surface of the painting, recto and verso. RÉSUMÉUne importante application de la spectro-imagerie ou imagerie hyperspectrale est la classification ou la différenciation de pigments en fonction des données de réflectance spectrale obtenues. Contrairement à un instrument d'analyse ponctuel, cette méthode non invasive examine la surface d'intérêt dans son ensemble. Cela signifie que son potentiel n'est pas seulement la différenciation de pigments mais aussi leur cartographie. Cependant, la difficulté réside dans le fait que dans une véritable peinture, il n'y a pas de limite nette entre des zones de pigments purs ou de différents mélanges de ces pigments. Les pigments et autres matériaux constitutifs d'une peinture se mélangent imperceptiblement et continuellement dans le domaine physique. Dans cet article nous présentons une approche basée sur la divergence de spectre pour la différenciation des pigments et leur cartographie. Cette méthodologie est ensuite appliquée au chef-d'oeuvre de Munch Le Cri (1893), dont les pigments et matériaux constitutifs ont été identifiés en plusieurs points de la peinture dans une étude précédente. Grâce à la méthodologie proposée, nous avons pu étendre les analyses ponctuelles de pigments et autres matériaux à l'ensemble de la surface de la peinture, recto et verso. Traduit par Claire Cuyaubère. RESUMOUma aplicação importante da espectroscopia de imagem ou imagem hiperespectral é a classificação ou discriminação de pigmentos com base na informação de refletância espectral obtida. Ao contrário de ser uma ferramenta de análise pontual, esse método não invasivo captura toda a superfície de interesse. Isso significa que seu potencial não está apenas na discriminação de pigmentos, mas também em seu mapeamento. No entanto, o desafio reside no fato de que, em uma pintura real, não há uma borda nítida entre regiões com certos pigmentos puros ou que contenham exatamente a mesma mistura. Pigmentos e outros materiais de pintura se misturam perfeitamente e continuam...

Quantitative analysis of mixed pigments for Chinese paintings using the improved method of ratio spectra derivative spectrophotometry based on mode

“…The color difference values dE of pigments samples was measured by the AvaSpec-ULS3648 series spectroradiometer produced by Avantes BV in the Netherlands, whose spectral range was from 360 to 780 nm. Difference operation was performed on the color difference value of adjacent pigments samples that can make it more intuitive to observe that how the color difference value (11)…”

“…Before extracting endmembers, it is sometimes necessary to perform component number estimation. The currently used methods include orthogonal projection, spectral clustering, geometrical and statistical approaches [7][8][9][10][11][12][13][14], which are selected to estimate the "true" or "pure" spectral components contained on the image, and some prior knowledge is usually required in some steps to eliminate possibly unimportant endmembers [8,13,14]. On the other hand, each pigment has its specific reflective spectrum.…”

Comparison of Imaging Models for Spectral Unmixing in Oil Painting

“…Moving a step forward, with the application of SU, it is possible to retrieve abundance maps, as gray-level images for example, considering the spectral signatures of the pure pigments as endmembers in the unmixing problem. Examples of recent works on PM have seen the inversion of the linear model, to map the pigments of Edvard Munch’s The Scream [ 19 , 22 ], and to perform pigment identification [ 23 ] using the sparse SUnSAL approach [ 24 ]. Recently, PM has been tackled with Deep Learning approaches as well [ 25 , 26 , 27 ].…”

Fully Constrained Least Squares Linear Spectral Unmixing of The Scream (Verso, 1893)

“…The use of spectral imaging for painting conservation started in the early 90s [3], aimed at providing more accurate documentation of cultural heritage paintings. In the recent decade, we can find works exploiting the use and advantages of HSI to provide, e.g., pigment and constituent maps, showing their presence [4] and also their estimated proportion or concentration in a mixture [5,6]. From a computational point of view, these works can be categorized as classification or unmixing tasks.…”