When we look with our eyes, we see the world around us; however, how we are able to interpret the world by looking at it is poorly understood.
In a study published in the October issue of Journal of Vision, Professor Ingo Fruend and undergraduate student Elee Stalker from York University’s Centre for Vision Research used artificial intelligence methods of image generation to compare visual perception of image manipulations.
A given assumption is that our system for visual perception has evolved to interpret the kinds of things that we see in the world. Although this idea is conceptually simple, it is surprisingly difficult to test rigorously, said Fruend. Previous testing has involved image manipulations such as blurring, which unfortunately degrades quality without changing much of the image content. The challenge is how to apply manipulations that change content, without unnatural distortion.
The manipulations used in the study “Human sensitivity to perturbations constrained by a model of the natural image manifold” either made images look less natural, or changed content of images while maintaining their natural appearance. Consistent with the idea that the human visual system performs well with natural images, the authors found that observers tended not to recognize images that had even small natural manipulations applied to them. In contrast, observers tended to easily recognize images, despite larger, but unnatural image manipulations.
In the study, the authors conclude that human tuning to natural images is more general than detecting deviations from natural appearance, and that humans have, to some extent, access to detailed interrelations between natural images.
The findings could benefit future studies that investigate such processes under more naturalistic conditions.
Fruend joined the Department of Psychology as an assistant professor in July 2017 and Stalker is an undergraduate student in psychology currently working in Fruend’s research group as a research assistant.