The Neurophysiology and Computational Mechanisms of Object Representation

Abstract

Listen

Translate article

Introduction A concise description of the representation of objects and faces provided by inferior temporal cortex neurons in the primate (macaque) brain is followed by new findings about how this representation operates in natural scenes and allows a number of objects and their relative spatial position in a scene to be encoded. Then a computational approach to how the object recognition processes described are performed in the primate brain is discussed as well as the types of strategy that the human brain uses to solve the enormous computational problem of invariant object recognition in complex natural scenes (Rolls 2008b; Rolls and Deco 2002; Rolls and Stringer 2006b). This contribution aims to provide a closely linked neurophysiological and computational approach to object recognition and categorization. Other approaches are represented in this volume and elsewhere (Biederman 1987; Fukushima 1989; Riesenhuber and Poggio 2000; Serre et al. 2007), and are compared with current approaches (Rolls 2008b; Rolls and Deco 2002; Rolls and Stringer 2006b). The Neurophysiology of Object Representation in the Inferior Temporal Visual Cortex Some properties of the hierarchical organization of the primate ventral visual system that lead to the inferior temporal visual cortex (IT), where object representations are present (Rolls 2000; Rolls 2007b, 2008b; Rolls and Deco 2002), are shown in Figure 14.1. The receptive fields of neurons become larger related to the convergence from stage to stage, and the representation develops over the stages from features such as bars and edges, to combinations of features such as combinations of lines or colors in intermediate stages such as V4 (Hegde and Van Essen 2000; Ito and Komatsu 2004) to objects in IT.