Humans perform object recognition effortlessly and accurately. However, it is unknown how the visual system copes with variations in objects' appearance and the environmental conditions. Previous studies have suggested that affine variations such as size and position are compensated for in the feed-forward sweep of visual information processing while feedback signals are needed for precise recognition when encountering non-affine variations such as pose and lighting. Yet, no empirical data exist to support this suggestion. We systematically investigated the impact of the above-mentioned affine and non-affine variations on the categorization performance of the feed-forward mechanisms of the human brain. For that purpose, we designed a backward-masking behavioral categorization paradigm as well as a passive viewing EEG recording experiment. On a set of varying stimuli, we found that the feed-forward visual pathways contributed more dominantly to the compensation of variations in size and position compared to lighting and pose. This was reflected in both the amplitude and the latency of the category separability indices obtained from the EEG signals. Using a feed-forward computational model of the ventral visual stream, we also confirmed a more dominant role for the feed-forward visual mechanisms of the brain in the compensation of affine variations. Taken together, our experimental results support the theory that non-affine variations such as pose and lighting may need top-down feedback information from higher areas such as IT and PFC for precise object recognition.
Keywords: EEG; computational model; feed-forward vision; invariant object recognition; psychophysics.
Copyright © 2017 IBRO. Published by Elsevier Ltd. All rights reserved.