Abstract
Most input to V1 is non-feedforward, originating from lateral and feedback connections. Using functional magnetic resonance imaging (fMRI) and multivariate pattern analysis (MVPA), Smith & Muckli (2010) showed that non-feedforward stimulated regions of V1 (i.e. responding to an occluded image quadrant) contain contextual information about the surrounding image, fed back from higher visual areas. In two experiments we investigated what features of the surround are necessary for meaningful feedback to non-stimulated V1. In Experiment 1, we studied how decreasing surrounding scene information affects feedback. Participants viewed two natural scenes where we modulated the visibility of the surround using a grey mask punctured with Gaussian bubbles of varying sizes, (Gosselin & Schyns, 2001). In Experiment 2, we investigated whether feedback signals carry information about the full configuration of the scene (global) or information close to the occluded quadrant (local). Participants viewed stimuli composed of four Gabors oriented at either 45°or 135°, one in each quadrant. There were four possible global structures: Right (all Gabors at 45°), Left (all 135°), Diamond, and X-shape. In both experiments, each stimulus was presented in feedback (occluded quadrant) and feedforward (corresponding quadrant visible) conditions. We then decoded the stimuli using V1 voxels relating to the lower right quadrant. Results showed that larger bubbles lead to better decoding, suggesting that a large amount of surrounding information needs to be visible for meaningful feedback to non-stimulated V1. Secondly, we found that feedback in the occluded quadrant mainly carries information about the local surround, rather than the global structure of the stimulus. Finally, both experiments showed that feedback signals from the surround interact with feedforward information in the quadrant.
Meeting abstract presented at VSS 2016