Low-level visual features correlate with distinct cortical regions during natural viewing of a movie

The cerebral cortex has predominantly been studied using abstract, simplified and often unnatural stimuli. We therefore wanted to characterize cortical function during natural viewing conditions, especially to determine the degree to which distinct cortical regions are differentially involved in the processing of distinct features. In a previous study, we used fMRI data from 8 human volunteers who freely viewed the first 25 min of a James Bond movie, and showed that the intensity variation over time of high-level features (faces, human bodies, language and color) as judged offline by separate human observers correlated linearly with the BOLD signal of specialized, partially segregated cortical regions (Bartels and Zeki, 2004a). To complement the previous study, here we used a computational analysis of the movie to determine the intensities of low-level features over time. These included global and local motion flow-fields, frame-by-frame pixel errors, contrast, luminance and color saturation. Parallel to our previous findings, our results showed that partially segregated regions of the visual system correlated with distinct low-level features, in accord with their preferences known from controlled experiments. The results also complement our findings from an independent component analysis of the same dataset (Bartels and Zeki, 2004b), in which a multitude of distinct cortical regions was segregated without a-priori hypotheses about their function, based entirely on their differential and specific functional response to the movie over time. In conjunction with our previous findings we conclude that during processing of dynamic, complex natural scenes distinct regions of the visual brain maintain a high degree of functional preference for distinct visual features