According to a classical, constructionist view, the later stages of the visual system assemble elementary inputs — like the oriented features encoded by V1 — into larger and more complex combinations, capturing the structural relationships that determine the visual world. But this approach has stumbled on the enigmatic second visual area, V2, whose neurons defy our intuitions about how to begin segmenting scenes and encoding the shapes of objects.

In this thesis we develop a framework for the study of intermediate visual processing in the primate, focused on computation and representation in area V2. Rather than try to predict the responses of visual neurons to arbitrary inputs, we test hypotheses about their function by generating targeted experimental stimuli. The stimuli we use reflect the messy statistical reality of natural images, rather than intuitions about object construction. We identify novel responses properties in macaque and human V2 that robustly differentiates it from V1. We propose mechanistic explanations for these properties by contextualizing them among existing models of hierarchical computation. And we link these properties to several perceptual capabilities — and limits — that appear to depend specifically on processing in V2, and imply striking consequences for everyday vision.