Woodcut Search

Our woodcut search feature uses Content Based Image Recognition (CBIR) computer vision technology to match the appearance of impressions made using the same or similar woodcuts on different broadsides. Woodcut reuse was common in the printing industry, and the same (or carefully carved duplicate) blocks were frequently reused in various contexts, sometimes over many decades. At times this reuse was random, but the themes depicted in the woodcut impressions were just as often carefully selected to enhance, contradict, and build narrative connections between different prints, books, etc. Additionally, many images and even poses became iconographic or rhetorical, immediately signaling special meaning to viewers—a fact of particular importance to print culture in an era of limited literacy.

EBBA’s woodcut search feature allows scholars to track this use and reuse of woodcuts and meaning-filled impressions through the entire ballad ecosystem, allowing scholars to find previously hidden connections and significance in the woodcut impressions themselves and in their interrelations to each other as well as to the texts and even tunes that they accompany. The system uses custom developed software to identify key “features” in images, such as the shape made by a curved or a textural pattern, and then classifies these using a fast, scalable, similarity clustering algorithm. As we add woodcuts to our collection, and as our CBIR technology makes new matches from those woodcuts, this software gradually learns and becomes more adept at identifying similarities across EBBA's ballads over time. For more information about this image recognition technology visit the Arch-V Git Repository.

Because of the “bag of features” system of impressions association employed, you will find that sometimes the numbers of “like” images returned to any one image you search might vary depending on your starting image. This is because not every family of images is identical; they are only “like.” If part of your starting cut is worn down or broken off, for instance, or a new feature is introduced into the cut in a recarving of it, this seemingly minor feature will influence your results. You might thus have to employ more than one image inroad to find all the sister images. You might also notice what appears at first a random “mismatch” in the results. In such cases, look carefully at where the feature lines in that "odd" image match those in the starting image; the computer may well be seeing connections that are, in fact there, and are logical to a machine-generated clustering algorithm, if not at first to the human eye.

Enjoy your woodcut impressions discoveries!