Early Modern Computer Vision
Leonardo Impett, M.A.
My research ambition is broadly to attempt to do for visual art what the Stanford Literary Lab has done for literary fiction: an approach which combines the computational analysis of 'big data' with traditional criticism and theory. Doing computation on images (looking for objects, gestures, iconographies) is inherently more difficult than text (recognizing word frequencies, classes, cases), but computer vision gives us the tools to close that gap. For instance, computer vision has enabled large-scale computational analysis of gesture and movement in image-databases, through the automatic recognition of body pose and gesture. I am currently building a larger-scale digital atlas of gesture, a platform for the diachronic study of both artistic and linguistic gestural evolution. Computer vision necessarily embodies a theory of vision (primarily a neuroscientific one). My current project, Early Modern Visual Computing, asks what computer graphics and computer vision might look like if they were based not on current theories of optics and vision, but on those of the European Renaissance; prototyping new methods that can be used as computational thought-experiments for the history of optical science and of the visual arts. Historical rendering algorithms, which implement pre-computational optical theories, can show that many presumed mistakes in optical science (e.g. Leonardo’s shadow-rays) can instead be read as heuristic rendering algorithms. This is particularly relevant in considering divine light, a common phenomenon of counter-reformation painting, which early modern treatises describe as physically distinct from solar or artificial light. This results in a form of Forensic Architecture: through a reconstruction of the 3D scene geometry (through depth estimation) and light sources (through computational sciography) implied in perspectival painting, the relationship between physical and divine light in a room can be charted.