Machine Learning for the Analysis of Artworks and Historical Documents

In recent years, a mass digitization of document images and artworks has taken place. In many cases, this makes an exhaustive search and traveling to archives/museums unnecessary but shifts the problem to the online domain. Automatic or semi-automatic methods make it possible to search, process, and analyze this big data in new ways. For example, the historian can get estimates of different document attributes, such as the dominant script type, the date, or the potential writer. These are related to different machine learning tasks: document classification, regression and retrieval. Similarly, in visual artwork analyses, image or object retrieval may relate different artworks to each other. This talk gives an overview over several computational humanities methods developed at the pattern recognition lab focusing on retrieval tasks.

Vincent Christlein studied Computer Science, graduated in 2012 and received his PhD (Dr.-Ing.) in 2018 from the Friedrich-Alexander University of Erlangen-Nürnberg (FAU), Germany. During his studies, he worked on automatic handwriting analysis with focus on writer identification and writer retrieval. Since 2018, he has been working as a research associate at the Pattern Recognition Lab (FAU) and was promoted to Academic Councilor in 2020. He heads the Computer Vision group, which covers a wide variance of topics, e.g. environmental projects such as glacier segmentation or solar cell crack recognition, but also computational humanities topics, such as document and art analysis.