Abstract
This paper presents an inferotemporal cortex model for object recognition and classification. The model is based on the modular network SOM, and produces a self-organized map of RBF networks. It can be trained to classify artificial 3D objects according to their structural similarity, and its properties are consistent with those in the TE area of the brain of macaque monkeys. We show that the model is able to distinguish between canonical and noncanonical views of objects. We further demonstrate that this ability leads to view-invariant recognition of objects by incorporating a mental-rotation type mechanism. Finally, we discuss the similarities and differences in view canonicality between the model and human observers.
