Corpus, corpora and visual methodologies
Since I am the graphics-person, I’m also collecting images, of course. I try to do as many screenshots as necessary and scrape the internet for additional material. There is a lot of paratext laying around, such as box art, magazin articles, illustrations and so on. This collection will sooner or later lead to a corpus of it’s own. My personal research archive, through which I do my visual analysis.
I started to use Tropy, an image application for archival research. It’s pretty easy to use, stable, and local first. It has a community and is made by the people who are also involved in Zotero and Omeka. So it won my heart pretty quickly. It enables me to build up my own ontology and describe my images as well as use tags and annotations.
Once in Tropy and slightly organised, I sometimes work a bit more freely in Miro. As an example I did a FAVR analysis and also tried to track some pop-cultural references.
This week, I also tried to apply some computer vision/machine
learning to the Ball Raider
collection. Through the Images
in Social Media Research conference, I got to know Memespector-GUI,
which is an interface to feed image-material into different
computer vision APIs and get the result as
json. I don’t know much about the open-source
models yet, so their results were meagre. The Google Vision API
result was pretty fine on the other hand. Have a look at the Ball Raider Google
Vision and VGG16 results. Here are some keywords for the
main character of Ball
Bodybuilder; Human body; Chest; Bodybuilding; Human anatomy; Nerve; Art; Font; Trunk; Abdomen
My corpus will grow, and I’d love to learn more on how approaches in the digital humanities can support me in my research. I could imagine that the corpus itself will become a scientific output of my research.