The art historic objective of VISTORY is to revive the study of stylistic development of painting with the help of computer learning. It will do so by developing a visual analytics framework in which we leverage the best of both worlds: Automatic analysis of large scale datasets by the computer, combined with the unique ability of art history experts to make detailed assessments of small sets of images. We will research how image features can be optimized by incorporating knowledge on the creation process and characteristics of materials.
From there, we will consider how to visualize the results in such a way that art historians can interactively annotate and re-group the paintings. Machine learning methods will then be developed which use those interactions to improve the model of painting similarity the system has. Integrating the resulting data-driven similarities among the paintings with existing structured metadata about the paintings yield our stylistic histories in terms of visual characteristics, place, and time. The resulting suite of tools will be made available to the community as open source software.