Proceedings of the Annual Meeting of the Cognitive Science Society

How does the visual representation of Chinese characters change with expertise? Fluent readers are known to perceive characters in terms of radicals and structural configurations, but the representational basis of perception before such knowledge is acquired remains unclear. We test whether novices' similarity judgments can be explained by generic shape computations from computer vision. Participants with three expertise levels rated pairwise visual similarity of Chinese characters. From pixelated character silhouettes, we computed 109 descriptors spanning multiple families of low-level shape statistics and derived pairwise distances. Using representational similarity analysis and Random Forest prediction, we compared descriptor distances with human judgments. Shape descriptors predicted novices' judgments strongly and learners' moderately but failed to predict fluent readers'. These findings establish a representational bridge between computer vision techniques and human visual cognition, suggesting that the generic shape geometry processing may form the starting point of visual similarity judgments before expertise introduces higher-level structure.