Object Size and Its Correlated Properties: Insights Into the Representation of Shape in the Ventral Visual Stream

Abstract

High-level visual cortex follows a large-scale organization related to the real-world size of objects, where some regions respond more to pictures of small inanimate objects while other regions respond more to large inanimate objects. In this thesis, I test the hypothesis that size, a visually inferred property, might be an organizational driver because it is meaningfully correlated with properties that are both more abstract and more vi- sual than size itself in nature. Specifically, I investigate the relationship between the real-world size of an object and its motor-relevance (Chapter 1) or its perceived curvature (Chapter 2). In both studies, I find that the real-world size organization is stronger and more reliable when responding to objects that follow the natural covariance with size (small and manipulable, or small and curvy), compared to objects that do not (big and manipulable, or big and curvy). These observations lend support to the hypothesis that the large-scale organization by real-world size is systematically related to a number of covarying properties. These covarying properties could either be behaviorally relevant (e.g. manipu- lability) or important features being encoded from earlier stages of visual processing (e.g. curvature). Different possible theories of why this organization arises are discussed. The thesis shifts its focus on Chapter 3 to the finer-grained similarity spaces of inanimate objects and how these reflected in responses to objects across the ventral stream. Specifically, I examined how neural responses to objects across the ventral stream hierarchy are related to three different behav- ioral spaces: object similarity measured implicitly by visual search, shape similarity as estimated by an explicit sorting task, and unguided object similarity estimated by the same explicit sorting task but without any instructions. I found that the object similarity space measured by visual search was well correlated with posterior regions of the ventral stream, and after a relatively sharp transition, the shape- based similarity space was well correlated with anterior regions of the ventral stream. Based on these data, I suggest that this relatively sharp transition along the ventral stream hierarchy reflects a transi- tion from more pictorial to more structural representation of inanimate object information. Overall, this thesis provides new theoretical contributions in the understanding the nature of the object representation along the ventral visual stream, suggesting that visual brain regions are sensitive to object properties at multiple levels of abstraction, related to both perceptual and action-related properties, with a transition from more pictorial to more structural levels of representation.