Abstract
Visual attention is traditionally thought to play a crucial role in binding of image features into complex visual representations. Recent findings from natural scene research challenge this idea, by showing that complex scenes can still be categorized under conditions of diminished attention. One possibility is that this preserved capacity is mediated by a 'gist' representation: a global scene impression that is formed even when the scene is outside the focus of attention. Here, we investigated to what extent the brain still processes global information when attention is diverted away from the scene. We presented several hundreds of natural scenes while subjects performed, in different blocks, a man-made vs. natural categorization task (full attention on the scene), a central letter task (focal attention diverted) and a peripheral outline task (distributed attention diverted). Simultaneously, EEG was recorded to compare ERP amplitudes for the same scenes presented under these different attention manipulations. At early occipital electrodes, evoked responses for man-made versus natural scenes differed reliably from 100 ms onwards, regardless of attentional set. At parietal-occipital electrodes, however, reliable differences appeared around 200 ms, but were only observed under full attention, not when focal or distributed attention was directed elsewhere. Using computational modeling, we show that the ERP differences are likely driven by differences between man-made and natural scenes in a biologically plausible image statistic called spatial coherence (SC) that reflects integrated local contrast across the entire scene (Groen et al., 2013, J Neurosci). Regression analysis of single-trial ERP amplitude on image statistics revealed a selective contribution of SC to evoked activity when a significant difference between man-made and natural scenes was present. Overall, these results suggest that the brain does extract global information under diminished attention, but that this information is only statistically integrated and broadcasted across the visual system under full attention.
Meeting abstract presented at VSS 2014