N300 sensitivity to statistical regularity persists for low-pass filtered scenes

Prior research showed that the amplitude of the N300 is smaller in response to “good” exemplars as compared to “bad” exemplars of natural scene categories (Kumar et al., 2021). We previously suggested that this N300 difference may stem from good exemplars providing a better match to an internal prediction. It has been suggested in the object recognition literature that initial coarse predictions about the identity of a visual stimulus may depend on a low spatial frequency (LSF) magnocellular signal (Bar et al., 2006). Here, we asked whether N300 amplitude differences for good and bad scene exemplars persist when only the LSFs are present in the scene. We recorded electroencephalography while presenting participants with good and bad greyscale exemplars of four natural scene categories (i.e., mountain, city, beach, and highway); half the images were filtered to leave only LSF information (low pass filtered). Participants were instructed to pay attention to the scene and then respond on a scale of 1 to 4 how much they liked the scene. We replicated Kumar et. al.’s (2021) findings for color images with the unfiltered greyscale ones, such that N300s were enhanced for bad exemplar scenes. Interestingly, we found a similar effect for the LSF scenes, indicating that good and bad scenes differ in LSF content. The fact that the good/bad N300 difference persists for LSF scenes is consistent with the idea that predictive coding signals may be partially carried by the magnocellular pathway.