Quantitative measures of crowding susceptibility in peripheral vision for large datasets

Abstract

Peripheral vision is simulated using several trained generative neural networks. These networks map an image to a synthesized mongrel. A mongrel is an image simulating the visual phenomenon of crowding that a normal human would experience in the periphery. Mongrels of natural scenes and font types are explored in this thesis. These synthesized mongrels and base images were scored by feature similarity to determine an images' quantitative susceptibility to the crowding phenomenon. The quantitative measure is used to determine the most and least susceptible fonts to crowding in a large data set of fonts.