Art, Vision, and Probability

Aaron Hertzmann
University of Toronto

Slides (PDF, 1 Mb)

Many modern theories of computational neuroscience view the brain --- or, at least, some parts of it --- as a probabilistic inference mechanism. Vision, in particular, entails converting visual stimuli into probabilistic neural encodings of scene interpretations, including both low-level and high-level features. Probabilistic reasoning seems to be essential to building highly-automatic computer vision systems as well.

Although no one *really* knows what's going on in the brain, I argue that such theories provide insights into the visual arts (including photography), and suggest how we might one day be able to build a computational neuroscience theory of art (or, at least, some aspects of art). Given such a model (or simplified versions of it, some of which are available today), we could use these models to build algorithms to help enable and improve artistic applications and photography.


Aaron Hertzmann is an Assistant Professor of Computer Science at University of Toronto. He received a BA in Computer Science and Art & Art History from Rice University in 1996, and an MS and PhD in Computer Science from New York University in 1998 and 2001, respectively. In the past, he has worked at University of Washington, Microsoft Research, Mitsubishi Electric Research Lab, Interval Research Corporation and at NEC Research Institute. In 2004, he co-chaired the NPAR conference on non-photorealistic rendering, served on the jury for the Annecy International Animated Film Festival, co-chaired a NIPS workshop on machine learning in vision and graphics, and was named to the TR100, a list of the top 100 technology innovators under 35 worldwide. His research interests include computer vision, computer animation, and machine learning.