Abstract
The normalization model of attention has been proposed as a unifying framework to account for various effects of attention in the visual cortex, and its success in predicting neural responses has been documented in primate electrophysiology studies. Here, using a human fMRI study, we investigated whether the normalization model can predict attentional modulations when participants attend to an object in a cluttered scene. We used a blocked-design paradigm in which half-transparent stimuli from the two categories of human bodies and houses were presented either in isolation or in pairs. A cue at the beginning of the block indicated the attended object. When paired, stimuli were superimposed to enforce object-based attention. We focused on the object-selective regions lateral occipital cortex (LOC) and posterior fusiform area (pFs), and the category-selective regions extrastriate body area (EBA) and parahippocampal place area (PPA) and determined the preferred and null stimuli for each voxel in each region. Results showed that shifting attention from the preferred to the null stimulus significantly reduced voxel responses in all these regions. Also, the effect of the unattended stimulus on the responses depended on voxel selectivity for that stimulus, with the unattended preferred stimulus having larger effects on the responses than the unattended null stimulus. We modeled voxel responses in different attentional conditions using a linear, a weighted average, and a normalization model. Results indicated that while the linear and the weighted average models were better than chance in predicting the responses, the normalization model had significantly better predictions than the other two models in all regions and especially captured the effect of voxel selectivity on the attentional modulations. These results suggest that when attending to objects in a cluttered scene, the responses in the object selective cortex are determined by divisive normalization.