Conference on Cognitive Computational Neuroscience

Recent studies have attempted to distinguish how visual features and semantic knowledge independently contribute to object recognition in the human brain. Here, we use images of real but unfamiliar animals, plants, and minerals to study how the perception of objects occurs in the absence of knowledge about their true category. We recorded brain activity with electroencephalography (EEG) while participants classified ambiguous stimuli as animals, plants and minerals, and during a fixation monitoring task. Results indicate that participants were largely naive of the true classes, and liberally classified objects as minerals but were more conservative for animals and plants. Despite poor classification performance, the true class of stimuli were distinguishable from multivariate patterns of neural activity in both classification and fixation monitoring tasks. Representations in the brain more closely matched behavioural classifications, but like true class, was best accounted for by a combination of mid-level features. Together, the findings suggest that mid-level features are sufficient for the brain to predict the true category of even deliberately confusing objects, and these features also explain behavioural similarity judgements.