Shape Perception in Human and Computer Vision by Sven J. Dickinson & Zygmunt Pizlo

Author:Sven J. Dickinson & Zygmunt Pizlo
Language: eng
Format: epub
Publisher: Springer London, London

18.1 Introduction

Our everyday perceptual experience is of a world populated by objects and surfaces arrayed in space, as well as of events that produce changes in these arrangements over time. Successful perception, thought and action depend on processes that produce accurate descriptions of these objects and events. Often, object contours are only partially visible as we move or as they move around us. Nevertheless, we experience a unified, stable world: the squirrel running through the tree branches appears as a single animal, not as dissociated squirrel-bits, and the house seen through the slats of a fence is one house, not a collection of independent house fragments. These perceptual outcomes depend on a number of segmentation, grouping, and interpolation processes, which, taken together, perform some of the most crucial and remarkable tasks in allowing us to perceive the world visually. They also pose some of the greatest challenges in understanding the underlying processes and mechanisms of vision.

Researchers in the past several decades have made considerable progress on a number of important components of these perceptual capabilities. Much is known about early cortical processing of visual information. At a more abstract level, experimental data and computational models have revealed a great deal about contour, object, and shape perception. Neurophysiological and imaging methods have provided evidence for functional specificity in areas of cortex for animate and inanimate objects, tools, faces, and places. However, between the initial encodings by spatially localized units and higher level descriptions of contours, surfaces, objects and their properties lies a considerable gap. To use a chess analogy, we do not understand much about the “middle game.” The study of shape perception and representation is important in its own right but also because it gives us a sharp focus on some of the biggest unsolved general issues in the computational and neural understanding of perception.

Early cortical encodings (e.g., responses of neural units in V1) are spatially local, retinally specific, and modulated by oriented contrast. The functionally important outputs of perceiving are constancy-based descriptions of bounded objects, their contours, surfaces, and shapes, and their arrangements in space. Our goal in this chapter is to shed light on shape perception, but also to use it as a vehicle to focus on major issues that must be addressed in order to understand how early visual processes connect to high-level representations. We describe (1) the dependence of shape perception on segmentation and grouping processes, and (2) properties that (some) shape representations must have and how they might be assembled from lower level encodings. In both discussions, we end with thoughts and efforts on a crucial frontier of work in these areas, which we might call “modeling the middle.”

Download

Copyright Disclaimer:
This site does not store any files on its server. We only index and link to content provided by other sites. Please contact the content providers to delete copyright contents if any and email us, we'll remove relevant links or contents immediately.