ARCA 2.0: Automatic Recognition of Color for Archaeology through a Web-Application


F.L.M. Milotta, C. Quattrocchi, F. Stanco, D. Tanasi, S. Pasquale, A.M. Gueli

Demo page for 2018 IEEE International Conference on Metrology for Archaeology and Cultural Heritage


Abstract



ARCA stands for Automatic Recognition of Color for Archaeology. Indeed, color specification is a common and critical issue for archaeologists, motivated by the need of identifying and cataloguing color names. For these purposes, Munsell Soil Color Charts (MSCCs) are largely employed. However, archaeologists are used to perform color specification by visual means, through a procedure that has been proved to be time consuming, subjective and error-prone. In its first version, ARCA has been presented as the outcome of our project for realizing an application which could be objective and easy to be used directly in the excavation sites. In this paper, we present ARCA 2.0: a totally reshaped method for color specification relying onto empiric transformations specifically tuned for cultural heritage color specification. As in the previous version, ARCA 2.0 is thought for RGB to HVC color conversion. We validate and assess the newer method, releasing a set of coefficients that can be used also by other researchers for color specification tasks. We also hereby present a demo of the web-application based onto experimental results reached during this research.

Experimental Setting



In this work, we employed Munsell Soils Color Charts (MSCCs) 2009 edition (excluding Gley1, Gley2 and White charts), counting a number of 331 chips. We remind here that MSCCs 2009 edition is currently the most complete tool used by archaeologists for soil color specification directly on-site. We acquired photos of MSCCs 2009 edition in a controlled environment. Charts were put in a Color Assessment Cabinet of VeriVide. Through the cabinet we were able to fix illuminant conditions to D65 (6500K, midday sunlight). We set acquiring devices within the cabinet, together with the MSCCs. We also acquired a Konica Minolta CS-A5 calibration plate, according to colorimetry best practice, in order to evaluate white normalization impact on our images. We acquired 10 images per device, counting a total of 331 Munsell chips per device. The color of each chip was manually sampled both in RGB and L*a*b* color spaces. The former has been conducted through our web-application, manually sampling the almost central pixel of each chip and computing the mode between its neighbours, in a patch of 49x49 pixels (2,401 pixels). L*a*b* sampling, instead, has been conducted with spectrophotometer Konica Minolta 2600d, carefully aligning the sampling device with each chip. L*a*b* values have been acquired for the validation phase: they will be compared to RGB sampled values converted to L*a*b*. Then, sampled L*a*b* and chip-printed HVC values are used as ground truth in our validation phase, while from RGB sampled triplets come L*a*b* and HVC observed values.

Dataset



Dataset is available under request. You may send an e-mail to one of the authors (corresponding author is preferred).


NOTE: A new extended dataset named ARCA1488 is available at http://iplab.dmi.unict.it/ARCA1488/

Method



Empiric transformations have been estimated starting from 8 polynomial models. Empiric transformations are based on an equation of this kind:

where Y are the expected (ground truth) HVC or L*a*b* values, while X are the observed ones. Once a method is chosen, then f(X) represents the polynomial terms of the related model, while t the coefficients. Y is a 3xS vector, t is a 3xN vector, and f(X) is NxS vector, where N is the number of polynomial coefficients in the chosen model, and S is the number of samples. Hence, given S samples, initially we estimated the coefficients t through least square method, then we validated t comparing predicted Y with the expected one, in terms of linear correlation (Pearson coefficient), CIEDE1976, CIEDE2000 and Godlove distances. Learned coefficients have been used for the implementation of ARCA 2.0 web-application. Major details can be found in the paper.

Conclusion and Results



Color specification employing Munsell Soil Color Charts (MSCCs) is a common and critical issue for archaeologists, motivated by the need of identifying and cataloguing color names. We presented ARCA 2.0: a method for color specification relying onto empiric transformations in which we defined a procedure for RGB to HVC color conversion. Validation has been performed with several metrics, showing that color conversion is reliable enough. Then, we released a set of coefficients to perform RGB to HVC color conversion through the best performing model among the many defined. Coefficients are available in the paper. Finally, we published a demo of the web-application based onto experimental results reached during this research.

References



This paper is part of the project ARCA. Previously published papers and datasets can be found in the followings: