Visual Crowding and Category-Specific Deficits: a Neural Network Model

Abstract

Various experimental approaches, using static 2D canonical views of living and non-living entities indicate that knowledge representations of these categories are distinct. In a series of experiments a Kohonen self organizing feature map was trained to recognise 2D digitised images. As a result, images of animals and musical instruments were represented within a shared set of processing units, which suggests that they are visually crowded categories, unlike clothing and furniture. These results are in keeping with those from other experimental approaches. Thus, it would appear that the simple interplay between a SOM and 2D images provides a valuable model of one route in visual object recognition.