Stains as Spectral Curves

Over the course of the last few months, the Library of Stains team has been analyzing data collected from the hundreds of images we gathered during three trips to the Library of Congress, the University of Pennsylvania, and the Universities of Wisconsin and Iowa.

Multispectral data analysis can take many shapes and forms. As part of the Library of Stains project, the team has applied a methodology specific to stains.  This post takes you through the steps of that process as it relates to characterizing stains and what is, and is not, possible to know.

The project uses Image J software, freely accessible on the internet, and the Paleo Toolbox, which was designed by Dr William (Bill) Christens-Barry of Equipoise Imaging.

The data that come to us straight from the camera images first need to be changed into .tif files.  This is done through Capture One, a Phase One software, and both sets of images can be saved in the same file. New software is being released soon and will do this step automatically.

Next, all .tif files need to be flattened.  Flattening images is done in two steps.  First you need to clean the flats – a series of shots of a white piece of paper taken on the same day as the images. Second, you flatten the set of images for a given side (or folio) to create an image that is evenly balanced agains the white light spectrum.  Using Image J again, the software does the work and automatically processes the series of “flats” in line with the manuscript images. The flattening process is a pretty easy learning curve, but somewhat time-consuming.  The good news is that this step will also be done automatically with the new software.

Once the files have been flattened, an intermediary step recreates a color image using Image J. This takes the flattened .tif files from a specific side (folio) and with the light information from the full stack of multispectral images, recreates a color image.  The Library of Stains team wanted to be able to have .jpg images that can be annotated and the specific areas on a given folio identified for the elements that were analyzed. 

Eventually the image will end up in Digital Mappa – a data curation software environment being used for our data visualization. In the meantime, to make our data analysis workflow as efficient as possible, we have put all the RGB color images into a powerpoint presentation.



Now, with the Powerpoint up on one computer, and Image J on another, we’re ready to begin analysis. We will be using  Image J once again to plot the z-axis for each spot that will serve to create a spectral curve.  For the the stain project, this meant we would be plotting a z-axis for the substrate (parchment or paper), the inks, the red and blue pigments used for rubrication and decoration, and of course, as many stains as our hearts desired.

To move from image files to spectral curves on an excel spreadsheet requires a few steps:

  1. Using the 10 non-filtered wavelengths we imaged for each side, we opened them in Image J and configured them into one stack.
  2. Scrolling through them highlights how different components on a given folio react to different light wavelengths.
  3. Plotting the z-axis on a particular part of the image is easy with Image J.  Outline the portion to be plotted with a small rectangle and choose “Plot z-axis.” The results immediately appear as a curve on the screen.  The curve can also be viewed as a list of number values.
  4. These values are plotted in the appropriate columns on an excel spreadsheet and labelled whichever element.  As shown below, we have columns for substrate, inks, pigments and stains.
  5. In order to see the true reflectance, the z-axis of specific component needs to be adjusted against the values for the white color checker.  Indeed, the white color checker values are the first to be plotted and inserted into the appropriate columns for each of the material components.
  6. Following the methodology devised by colleagues at the Preservation and Testing Division of the Library of Congress, on the spreadsheet the formula is automatically calculated by dividing the intensity of the ink, substrate, pigment or stain by the intensity of the white swatch on the color checker. This then is what is plotted on the x and y axis of the spectral curve – the x axis showing light wavelengths, and the y axis, reflectance levels.

Then comes the fun part.  We are able to begin to decipher the curves.

Spectral Curves for inks stains on University of Wisconsin manuscript MS 170A, no. 8.
Spectral Curves for all blue inks found in the University of Wisconsin manuscripts.
Spectral curves for all red inks found in the University of Iowa manuscripts.
Spectral Curves for possible wax stains in the University of Iowa manuscripts.
















Preliminary results from the University of Iowa and Wisconsin show variations of ink curves, as well as possible curves on a number of folios that may indicate wax residues.  Further analysis is underway and final results, alongside the data itself, will be ready for open access by August 31, 2018.

With much thanks to Leah Pope Parker, PhD candidate in English at the University of Wisconsin, for her intellectual contributions to this project, as well as countless hours analyzing and visualizing data.


And the journey begins

20171106_113058 (1).jpgLast week we all convened in sunny Philadelphia to begin imaging stains from the Chemical Heritage Foundation and Penn Libraries manuscript collections.

With the generous help of Mike Toth from R.B. Toth Associates and Sarah Reidell, Margy E. Meyerson Head of Conservation of the Kislak Center for Special Collections, Rare Books and Manuscripts, the multispectral imaging system was set-up in a small windowless–i.e. perfect for imaging!–room within the Conservation Studio.

System setup
System setup
setting up a manuscript for imaging
Alberto Campagnolo and Sarah Reidell carefully setting up a manuscript for imaging
Heather Wacha, Erin Connelly, and Alberto Campagnolo setting up a manuscript

Over the course of two intense days we imaged stains from the pages and covers of fourteen manuscripts ranging from the 13th to the 16th century, thus beginning to build our dataset of stains. The manuscripts include nine alchemy texts from the Othmer collection, Chemical Heritage Foundation (Othmer MS 1 pictured below) and five medical texts from the Schoenberg Institute for Manuscript Studies and Penn Libraries collection.

We used a Phase One IQ260 Achromatic camera, a 60 megapixel 16-bit monochrome digital back with 8964 x 6716 pixel CCD array at 6.0 micron pixel size, with an iXR body and 80mm lens producing 675 ppi resolution images. The special illumination necessary for multi-spectral imaging was provided by a third-generation LED light system designed by Dr. William (Bill) Christens-Barry of Equipoise Imaging that produces very specific and narrow bands of illumination, ranging from ultraviolet light (370nm) to the near infrared (940nm).1 Because of the nature of the project, we also utilized long-pass green and red filters to detect fluorescence energy: the filters remove the illumination wavelength, but let through longer fluorescence emission that can be recorded in the captured image, thus allowing the characteristic spectra of substrate, colourant, or contaminant substances to be more completely determined and analyzed.
The camera-light-filter system is integrated within a software that simplifies the operation and records unified metadata at each step.

The result of the imaging is a sequence of photographs, one for each  different illumination and filter setting, as it can be seen below.

Image stack
Example of image stack for Left cover of Othmer 1, CHF
2017-11-16_16h56_46 (
Animation of image stack for Left cover of Othmer 1, CHF

Different materials react differently to each wavelength, and details that are not visible in natural light begin to appear and be clearly noticeable. Notice, for example, how the stain in the cover above appears and disappears, depending on the illumination.
One detail of particular interest is a writing in the upper part of the cover that was almost invisible to the naked eye, but that becomes immediately distinguishable and readable under infrared light (see detail images below).

Writing on Left cover of Othmer 1 (CHF)
Detail of writing on Left cover of Othmer 1 (CHF)

Capturing the photographs (and managing the metadata) is only the first step. For a deeper understanding of the data recorded and the variety of material responses to the different wavelengths, one needs to process the stack of images and analyze the data through statistical algorithms capable of simplifying it and of finding patterns in it.
This kind of analysis, thanks to colour reference cards positioned in the scene, can also reconstruct colour images, despite the fact that the camera is achromatic, i.e. agnostic to colour information (see below).

Reconstructed colour photograph of Othmer 1 (CHF)
Reconstructed colour photograph of Othmer 1 (CHF)

One output than can prove particularly useful in distinguishing different components — i.e. materials reacting in different ways under the different lights — is a false colour image, where different components are assigned an arbitrary colour to help discerning similar and dissimilar light responses.

False colour detail of Othmer 1 (CHF)
False colour detail of Othmer 1 (CHF)

It is through this kind of data analysis that we’ll try to distinguish and characterize stains in the coming months.

We thank Mike Toth, Bill Christens-Barry, James (Jim) Voelkel, William (Will) Noel, Doug Emery, and Sarah Reidell and everyone else involved with our imaging session at the University of Pennsylvania for their help and support.
We thank CLIR for their constant assistance (above and beyond financial support) and encouragement.

The team
The team: (from the top left) Erin Connelly, Mike Toth, Bill Christens-Barry, Heather Wacha, Alberto Campagnolo

1. We imaged at: 370nm (UV); 448nm (deep blue); 476nm (blue); 499nm (cyan); 519nm (green); 598nm (amber); 636nm (red); 740nm (IR1); 850nm (IR2); 940nm (IR3). UV in italics, visible light in roman characters, and infrared frequencies in bold.