CSAIL Event Calendar

Thesis Defense: Ambiguous Statistics -- How a Statistical Encoding in the Periphery Affects Perception

Recent understanding in human vision suggests that the periphery compresses visual information to a set of summary statistics. Some visual information is robust to this lossy compression, but others like spatial location and phase are not perfectly represented, leading to ambiguous interpretations of the statistical representation. Using the statistical encoding, we can visualize the information available in the periphery to gain intuitions about user interface design, or more generally, whether the periphery encodes sufficient information to perform a task without additional eye movements.

The periphery is most of the visual field. If it undergoes these losses of information, then surely our perception and ability to perform tasks efficiently are affected. We show that the statistical encoding explains human performance in classic visual search experiments. Based on the statistical understanding, we also propose a quantitative model that can estimate the average number of fixations humans would need to find a target in a search display. Further, we show that the ambiguities in the peripheral representation predict many aspects of some illusions. In particular, the model correctly predicts how polarity and width affects the Pinna-Gregory illusion. Visualizing the statistical representation of the illusion shows that many qualitative aspects of the illusion are captured by the statistical ambiguities.

See other events happening in September 2012