Abstract
Previous research has shown that the perceptual similarity between object shapes is a major determinant of classification and discrimination performance (Wasserman & Astley, 1994). Behavioral studies showed that the human visual system represents these perceptual similarities in a faithful way (e.g., Edelman, 1998). We tested the representation of similarities between parametrically varied two-dimensional shapes (Fourier Descriptors) in the inferotemporal cortex of two awake rhesus monkeys. The shapes were divided into three ‘categories’ with distinct radial frequency components, resulting in qualitative shape differences. Within each category, metric differences between shapes were induced by varying independently the amplitude of two radial frequency components in effect creating 8 shapes for each category arranged in a square-like manner in the two-dimensional amplitude space. We recorded from 75 neurons while the monkeys were fixating and discriminating the stimuli. A similarity matrix was constructed by calculating the similarity between each stimulus combination in the multi-dimensional space spanned by these 75 neurons. The application of cluster analysis and multidimensional scaling to these data showed that the representation of the similarities between shapes in the inferotemporal cortex is highly veridical: both the clustering into three categories and the within-category shape arrangement were reflected in the neural similarities. While the responses of many neurons were determined by category membership, there was a surprisingly large proportion of neurons with similar maximum responses for different categories (no clear between-category selectivity), but with good within-category selectivity.