As this is the first post on the brand-new Seebibyte blog, I’m taking the liberty of showcasing some of my own research on historical printing.

My introduction to what we now call VISE came as part of the 2011-12 Bodleian Broadside Ballads Online project. Based on Relja Arandjelović’s doctoral work, Bodleian Ballads applied computer vision to printed illustrations through a tool branded as ImageMatch.

As documented in this publication, the tool is intended to find matching woodcuts in early English broadside ballads. In the example shown below, two woodcuts on different ballad sheets are matched with a high degree of probability.

Ballads match

Some of the woodblocks used to print ballad illustrations survive: could VISE also match their impressions from a query image of the block? It turned out that it could, but only after a few adjustments to the block image.

For this experiment I created a VISE Search Engine consisting only of images of one block and several of its known impressions. This block, formerly employed in Newcastle-upon-Tyne by several generations of printers and later formed into an important collection, is now in the British Museum.

Hind block

An untreated image of the block returns a low score against a known impression: the features that are matched are in most cases spurious and would probably be disregarded as noise in any real-world search scenario.

Hind original

I then applied some simple transformations in an image editor: I used Photoshop, but GIMP or other tools would work just as well. By far the most important transformation was to horizontally flip the image so that it matched the orientation of the printed impression. VISE tolerates rotational changes, scaling and skewing but since it attemtps to match ordinary objects in the world it does not search for their reflections, as might appear in a mirror or, in the case of a woodblock, an offprint. A future version could easily do so, but for now print researchers must flip their query-images.

Hind block flipped

In order to give VISE the best possible chance to make a match I also converted the image to greyscale; inverted it so that dark became light and vice-versa; and then applied a combination of the Threshold and Levels filters to increase the contrasts between light and dark regions.

Photoshopped Hind blockflipped

It is not always easy for VISE to distinguish between the printable surface of a relief woodblock and the sub-surface, but enough contrast exists in this block around the edges of the raised surface to allow matching, which is now greatly improved.

processed matched

Many museums and other institutions hold unidentified woodblocks and other printing surfaces: coupled with the increasing availability of online images of publications, VISE provides an entirely new means of identification.

Giles Bergel