Munsell Color Specification
using ARCA (Automatic Recognition of Color for Archaeology)
Journal on Computing and Cultural Heritage (ACM JOCCH), 2017
1University of Catania, Department of Mathematics and Computer Science - Italy, {milotta, fstanco}@dmi.unict.it
2University of South Florida, Center for Virtualization and Applied Spatial Technologies (CVAST) & Department of History - Florida, USA, milotta@mail.usf.edu, dtanasi@usf.edu
3University of Catania, Department of Physics and Astronomy - Italy, agueli@unict.it
ARCA328 Image Dataset
NOTE: A new extended dataset named ARCA1488 is available at http://iplab.dmi.unict.it/ARCA1488/
==> Download from here the ARCA328 Image Dataset. (Last Updated 12/04/17)
Abstract
Munsell Soil Charts are a very common tool used by archaeologists for the color specification task. Charts are usually employed directly on cultural heritage sites to identify color of soils and collected artifacts. However, charts are designed to be used specifying the color through subjective perception of users, by visual mean, in a time consuming and error-prone procedure. It is likely that two users may estimate different Munsell notations for the same specimen, as colors are not perceived uniformly by different people. Hence, estimation process should be repeated several times and by more than a single expert user in order to be considered reliable. In this work, we employ our framework ARCA: Automatic Recognition of Color for Archaeology, specifically designed to provide a method for objective, deterministic, fast, and automatic Munsell estimation. ARCA is a valuable asset for archaeologists as it provides the definition of a smooth pipeline for an affordable Munsell notation estimation: image acquisition of specimens with general purpose digital cameras in an uncontrolled environment, manual sampling of specimen images in the ARCA desktop application, automatic Munsell color specification, and report generation.
We further assess our method with improved color tolerance validations and evaluations, introducing a comparison between Delta E 2000, Delta E 1976, Delta L*, Delta a*, and Delta b* differences. One of the main contribution of this paper is the extension of our former dataset ARCA108. We gathered two additional sets of images obtaining a new dataset consisting of pictures of Munsell Soil Charts Editions 2000 and 2009 plus images from a real test-case with 16 pottery shards. The new dataset counts 56,160 samples and 328 images, so it has been called ARCA328. Experimental results are reported to investigate which could be the best configuration to be used in the acquisition phase.
Dataset Acquisition
We acquired three sets of images, in the same place and hours, but different days, all almost sunny with some cloud cover. Our acquisition was set in Tampa, Florida (US), in GPS coords 28°03'47.9"N 82°24'40.9"W, and it was performed from 10:30 am to 12:30 pm with an unguided approach. Hence, we did not applied fixed positions or known angles of view for cameras. A Gretag-Macbeth color checker has been also employed in some experimental setting, in order to evaluate how the Munsell notation estimation may be influenced by the presence of color references.
The sets of images can be detailed as follows:
- Set 1 - ARCA108: This set contains images of a Munsell Soil Color Charts (Year 2000 Revised Washable Edition). This edition counts 9 sheets, namely: 10R, 2.5YR, 5YR, 7.5YR, 10YR, 2.5Y, 5Y, GLEY1, GLEY2. This set was acquired on March 8, 2017.
- Set 2: This set contains images of a Munsell Soil Color Charts (Year 2009 Edition, often referred as Tropical Soils Charts). This edition counts 13 sheets, namely: 5R, 7.5R, 10R, 2.5YR, 5YR, 7.5YR, 10YR, 2.5Y, 5Y, 10Y-5GY, GLEY1, GLEY2, WHITE. This set was acquired on April 14 and 17, 2017.
- Set 3 : This set contains images of 16 pottery shards. Fragments come from real excavation sites. For this reason they may be used as a real test-case for our ARCA application. This set was acquired on June 1. 2017.
We acquired the sets with the following possible settings:
- 2 kinds of devices: professional DSLR (Digital Single Reflex Camera) and common smartphone;
- 3 automatic white balancing algorithms (executed by the devices in the image capture phase): automatic, sunny (corresponding to standard illuminant D65: 6,500 K) and cloudy (corresponding to standard illuminant D75: 7,500 K);
- 1 fluorescence presetting: direct sunlight;
- 1 ISO setting: 400 ISO;
- 1 focus setting: autofocus;
- 2 kind of subject: the chart itself and the chart with a Gretag-Macbeth color checker nearby.
In this way, we obtained a total of 12 configurations for each Munsell chart. The resolution of the images is 5184 X 3456 pixels and 3840 X 2160 pixels for pictures taken by a DSLR camera and smartphone, respectively. All the images were saved in the standard JPG format, with a lossless setting for the quality (the highest possible).