Conference on Cognitive Computational Neuroscience

Visual object perception involves detecting complex features in a particular spatial configuration. It is widely believed that object perception happens largely independent of cognition and is supported by representations in the ventral visual cortex (VVC). In previous findings, we have shown to the contrary that although human VVC contains rich featural representations capable of discriminating objects of different categories, it does not explicitly encode the natural configuration of features that defines an object and therefore is a texture-like, not object-like, representation. Here, we extend these results to demonstrate that attention integrates visual features in their spatial configuration to generate object-like representations. Using deep texture synthesis, we synthesized stimuli ("synths") which contain the same complex visual features as natural images in scrambled configurations. We manipulated covert attention while subjects discriminated natural images of objects from their corresponding feature-matched synths. We found that focal spatial attention improved subjects' ability to discriminate natural objects from feature-matched scrambles. Furthermore, using BOLD imaging, we found that the neural representations in the ventral visual cortex became more discriminative between natural objects and feature-matched scrambles. Taken together, these preliminary results suggest that attention plays an important role in integrating complex visual features in their natural spatial configuration to support object perception.