Construction of Objects from Pixels — Cortical Coding in Intermediate-level Visual Area V4

Abstract

A fundamental role of the visual system is to transform pixel-based retinal images into the representation of objects. A key to understand this transformation is in intermediate-level cortical areas where the features extracted from an image are selectively pooled for the construction of "proto-objects." This short review introduces recent findings on intermediate-level cortical representation and discusses their essence towards understanding the abstraction of retinal images into objects. Recent physiological studies have reported that a population of neurons in intermediate-level visual cortex, V4, are capable of predicting figure and ground (FG) regions that provide the basis in the construction of a proto-object by defining its extent. The agreement in the characteristics of FG determination between the neural responses and perception strongly supports the crucial role of V4 neurons in FG perception. An important factor for the construction of a proto-object is its contour shape. Specifically, how the neurons in intermediate-level cortex code complex contours in natural scenes provides important clues for understanding the construction of proto-objects. A recent study reported that V4 neurons simultaneously code multiple contour features such as closure, curvature, orientation, and symmetry, which appears to be effective for the representation of complex contours. The result also raises a fundamental question on the representation of shape; whether they are coded by a combination of the geometric features such as curvature and orientation or by some other nonintuitive coding. These findings indicate a crucial role of a population of V4 neurons in the construction of proto-objects wherein FG regions are estimated and contour components are integrated.