The dominant evidence that supported this view for years has been a reversed photograph of the ink that seeped through to the verso side of the page, where the characters were not obstructed by the printed decoration. The visual appearance of the signature on the verso side is shown, here reversed so that it reads in the legible direction:
Ink showing through from recto to verso side of leaf (image has been reversed left to right)
The value of clarifying the signature on both sides of the page arguably is self-evident. The proposal to image this page was made by Mr. Andrew Henning, a senior in the Sally Barksdale Honors College of the University of Mississippi, as a result of his senior thesis. The trip was organized by Dr. Gregory Heyworth, Associate Professor of English and Head of the Lazarus Project, whose mission is to image culturally important artifacts. Team members included two other students in the Honors College, Kristen Vise and Mitchell Hobbs, Robert Jordan (photographer), Dr. William A. Christens-Barry (Chief Technical Officer of Equipoise Imaging LLC in Ellicott City MD, who designed the spectral illumination panels), and myself.
With the support of Renate Mesmer, the Folger’s Head of Conservation, and the staff of the Library, our team brought a scientific digital camera, spectral illumination panels, and processing computers to the Library on 12-13 March to collect and process images of the front and back of this page. This effort is similar to work previously performed by members of the team on a variety of manuscripts and printed material, including the Archimedes Palimpsest , the 1507 Universalis Cosmographica by Martin Waldseemüller at the Library of Congress, the David Livingstone Nyangwe Diary, and palimpsests at St. Catherine’s Monastery. During our work at the Folger, we hoped at minimum to contribute new evidence that could be assessed by scholars about the authenticity of the signature. If successful, this might also help the Folger assess the value of spectral imaging for use with other books and manuscripts in the collection.
The imaging system illuminated the object with 12 bands of light generated by light-emitting diodes (LEDs) with dominant wavelengths ranging from the near-ultraviolet region (just shorter than the range of visibility of the eye), through the visible spectrum, and into the near-infrared region of the spectrum at wavelengths longer than the eye can see. The digital camera used a 39-megapixel monochrome digital sensor (7216 × 5412 pixels), which has the distinct advantage over a color sensor of utilizing all light reflected from the paper, thus minimizing the exposure of the fragile object.
Since the collection of images just finished as I write these words, we have implemented only the most preliminary processing of the data, but the images provide encouraging results. The processing demonstrated here is based upon the statistics of the pixels over all 12 collected bands. These statistics are used to construct an equivalent set of 12 images from weighted sums and differences of the original bands. The first processed monochrome image includes the widest possible range of contrast over the original 12 bands of data and often resembles the black and white version of a color TV image. The second band shows the widest range of image contrast that is independent of the first. The process is repeated until 12 bands are constructed. Since the highest-order bands include the smallest range of contrast, they often include random variations due to noise. After the processing is completed, objects of similar “color” tend to be grouped together in the processed images, which may allow segmentation of overlapping objects whose appearances differ only in subtle ways. The bands produced from this statistical process are the principal components and the algorithm is principal component analysis.
The first seven (of 12) principal component bands of the front side of the page are shown below; note that the faded signature is readily visible, and separated from the printed decoration, in bands 4, 5, and 6. Some details are more obvious in this image than on the image from the verso side, such as the letter “m” following the flourish of the “W” which is not visible in the reversed image from the verso side.
First 7 (of 12) principal components calculated from the 12 bands of spectral data. Note that the signature is readily visible in bands 3 through 6
To further assist examination of the text, different image bands may be selected for rendering in “pseudocolor,” which then may be varied dynamically by changing the display parameters. This is most usefully performed by the user in real time, but the animation gives some feeling for how the visibility of the text varies with the rendering. Note that different features of the handwritten signature may be more readily visible at different times in the image sequence.
Animation to illustrate variation of the pseudocolor rendering generated from the principal components
We will be very interested to hear any reactions from scholars about the value of these images; these comments will help us improve and adapt the processing. We also look forward to future collaborations with the Library.