The preservation of cultural heritage and treatment thereof are delicate responsibilities that demand the best possible technologies and extreme care to avoid any irreversible loss. It necessitates a deep understanding of constituent materials, along with the analytical methods and cutting-edge technologies. Considering this direction, the goal of this dissertation is to promote the conservation procedures by providing an applicable workflow for spectral-based pigment identification. The proposed pipeline is a novel and practical aid for museum conservators for many aspects, such as inpainting, treatment and archiving of artwork.
Spectral-based pigment identification algorithms rely on accurate spectral data, a subtractive mixing model and an effective unmixing algorithm. In this dissertation, the spectral data were obtained using a spectral image acquisition system as a feasible and non-destructive technique. A liquid-crystal tunable filter (LCTF) and a CCD camera were used for spectral measurement of the painting. The spectral accuracy and precision of the LCTF-based spectral acquisition system were assessed and enhanced. Of the common factors affecting the acquisition performance, capturing geometry, LCTF angular dependencies and spectral characterization algorithm were new contributions to the traditional workflow.
The complexity of subtractive mixtures limits the effective application of linear unmixing algorithms for pigment identification. Accordingly, a new linear modification of single-constant Kubelka-Munk theory was derived to enable the use of available linear spectral unmixing algorithms for paint mixtures. A selection of geometric and iterative-based unmixing algorithms was applied to the LCTF spectral images in the subtractive mixing space using the defined subtractive linear model. Final sets of primary pigments were improved employing a pre-existing database of common pigments as a tool for practical inpainting procedures. The pigment maps, showing the concentration of each pigment per pixel, and RMS error images were calculated after estimating the primary pigments. The estimation errors and the performance of the pigment selection algorithms were assessed using 23 validation paintings.
Finally, the primary pigment information combined with multi-spectral images from a commercial 6-channel imaging system was used for visualization and high-resolution spectral image reproductions.
Library of Congress Subject Headings
Art--Reproduction--Data processing; Image reconstruction; Colorimetric analysis; Spectrophotometry
Color Science (Ph.D.)
Roy S. Berns
Mark D. Fairchild
Abed, Farhad Moghareh, "Pigment Identification of Paintings Based on Kubelka-‐Munk Theory and Spectral Images" (2014). Thesis. Rochester Institute of Technology. Accessed from
RIT – Main Campus