New computing tools could save graduate students from thousands of hours spent visually inspecting historical maps, quilts, medieval art, and manuscripts.
Humanities researchers use laborious visual inspection of historical materials to determine authorship and artistic lineage for manuscripts, maps, and quilts. A typical research project could take the time of several graduate students two or more years.
“We are interested in understanding how to support these domain scientists so they are more efficient,” said Peter Bajcsy, principal investigator for the Digging into Image Data to Answer Authorship-Related Questions (DID-ARQ) project at the National Center for Supercomputing Applications. He added, “We are giving them tools that will automate the visual inspection process.”
Those tools are the product of extensive collaboration between humanities researchers and Bajcsy’s team. To create them, the computer experts engaged in lengthy discussions to determine what questions the humanities researchers seek to answer, and exactly how they would answer them using traditional methods. Their inquiries sought out details such as which image features researchers use – consciously or unconsciously – to identify the author or artist. Then the computer scientists developed algorithms that could rapidly identify those features automatically.
“It’s a discussion that leads to solving a mutually interesting problem,” Bajcsy said. “For us, the interesting part is automation and computational scalability.”
Acquiring funding to develop the appropriate scalable algorithms was not easy. The biggest challenge, Bajcsy explained, was to identify a common theme that could bring together several humanities researchers focusing on very different historical artifacts. For DID-ARQ, the common theme turned out to be finding authorship of historical artifacts through image analysis.
Working together, researchers from NCSA, University of Illinois at Urbana-Champaign, Michigan State University, and the University of Sheffield were able to find three sets of data that could be analyzed by similar algorithms: 15th –century manuscripts, 17th and 18th –century maps, and 19th and 20th –century quilts. By forming a collaboration that promised insight into three large data sets, each team member was able to receive the necessary funding from the United States (National Science Foundation and the National Endowment for the Humanities) and the United Kingdom (JISC).
“Hopefully by doing this interdisciplinary research, there is going to be a seminal work in terms of our statistically significant understanding of authorships,” Bajcsy said, “because before now nobody has been able to look at a thousand images.”
Want more technical details? Listen to Bajcsy discuss DID-ARQ and historical maps. (Note that this is the same audio clip that is linked above.)
—Miriam Boon, iSGTW