Digital Pigment: From colour in Cultural Heritage to industrial requirement in spectral metrology – DigiPi

Our ambition is to create the foundations of spectral metrology by:
- A robust and accurate measure of the distance between two reflectance or radiance spectra,
- The characterisation of the spatial variations of reflectance spectra acquired from a coloured surface perceived as homogeneous but which is spectrally not uniform.

The necessity to control the stability and the accuracy of such optical measures is always required in civil industry, e.g. in military, Aerospatiale, medical applications, or in the valorisation of the cultural heritage. There is no metrology solution corresponding to this need while the use of spectral sensor increases. This lack is related to the problem complexity: measure processing is related to Digital Sciences, but the validation and interpretation comes from Physics. DigiPi proposes to break this boundary by combining two laboratories, one expert in digital processing of colour and the other one expert in the characterisation of complex heterogeneous materials and modelling their optical properties.

As part of this bi-disciplinary approach, we will produce new expressions of similarity/distance measure between spectra embedding constraints of robustness, genericity and uncertainty reduction. Our contribution will integrate the mathematical specificity of the spectrum, which is not a vector or a probability density function, but closer to a series or a function. Another contribution of DigiPi will decompose the distance measure in subparts allowing straightforward interpretations well reflecting the spectral variability.

There is no metrology without standards or references. We will create surface ranges ordered by the change in different manufacturing settings. The characterisation by various methods (profilometry, microscopy, spectrophotometry ...), and physical modelling of the optical properties of the paint layers will allow us to better understand the spectral variability of a pigment according to the physical and chemical characteristics. In particular, we will adapt existing models to embed parameters never previously integrated. All this knowledge will allow us to measure the rank correlation between the spectral distance measurements and the data characterisation or parameter models to obtain a complete validation. These surfaces and reference data will allow the development of this metrology beyond DigiPi.

The second objective is to produce digital attributes characterising non-uniformity colour from hyperspectral images. This objective is related to the analysis of the micro-textured surface appearance. Our scientific contribution will be to define digital attributes that must be metrologically valid in the physical sense (spectral aspect) and that can be correlated with models of human vision (colour and subjective aspect). The digital attributes will be compared using rank correlations to morphological and chemical parameters extracted from the physical models describing the paint layers of reference surfaces. In parallel, fractal image models coming from the work of the CIE TC8-14 will be used to assess the uncertainty stability whatever is the spatio-chromatic content.

Distance/similarity measures between spectra and attributes characterising the non-uniform appearance correspond to industrial, economic, medical, environmental ... requirements. The last part of DigiPi will establish the performance gain in two use cases: one related to industrial quality control in production of colour products and the other in cultural heritage context to characterise colour shading-off in a wide collection of royal vellums.