Image Matrices: Learning from Klosterneuburg
Speaker: Maximillian Schich, Northeastern University. Moderator: Noel Jackson, MIT. Abstract: This talk introduces image matrices as a useful visualization method for visual classification networks, such as tagged images or otherwise categorized visual documents. Famous examples of image matrices include the Verdun altarpiece at Klosterneuburg (1181), and the schedule of Las Vegas Strip hotels in Learning from Las Vegas (1972). Today image matrices are increasingly common in biology to present dynamic processes in cells, or to validate homologic statements in phylogenetic morphology. Going beyond the history of the image matrix, this talk explains how image matrices facilitate a unified approach of archeology and art history, connecting quantity and quality in a single workflow. Regarding data we will take a look at large classification networks of visual documents depicting ancient monuments, i.e. subjects of art history connected to subjects of archaeology via thousands of visual classification links. Zooming in from network diagrams of entire datasets to individual objects, we will see how image matrices can be used to uncover hidden mesoscale relationships that lie beyond the cognitive limit of an individual reading literature or using traditional database interfaces. In a particular example regarding imperial baths in Rome we will see, how image matrices can help to quickly extract qualitative stories from large datasets, that are neither explicit in the written record nor in the image classification at a granular level. The presented method is a subject of the author's Ph.D., a Max-Planck patent application and a project funded by the Innovation Fund of the President of Max-Planck-Society.