Abstract
Many studies suggest 3D objects are represented by sets of 2D views in recent years. However, it is not clear what features the views have. In this study, we show that 3D objects are represented in brain by a few canonical views spontaneously selected. It is also shown that canonical views correspond to ‘generic view’ which is robust against transformation by rotation in depth, by analyzing similarities in each view of objects with multi-dimensional scaling. Furthermore, comparing configuration of views in principal component space with the one in similarity space, we propose that it is imperfect to explain human recognition space with only principal component space because symmetry of shape is greatly related to object recognition.
