Purchase this article with an account.
Rudiger Heydt, Fangtu T. Qiu, Todd J. Macuda; Border ownership coding: global structure in local feature maps. Journal of Vision 2002;2(7):712. doi: 10.1167/2.7.712.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
The distinction between effects of stimulation inside and outside the “classical receptive field” has become an important concept in studying visual cortical neurons. We will present examples of this seemingly confusing mixture of local and global information in single cells and discuss how it might be used in the system. Feature selective cells of area V2 have small “minimum response fields” (outside which an optimally oriented contrast border does not produce a response), which means that they signal the precise localization of features. However, many cells show also strong modulation of responses depending on whether the local contrast border is part of a figure on one or the other side of the receptive field (Zhou et al, J Neurosci 2000). This “border ownership effect” can be produced by stimulus features far outside the classical receptive field — assigning borders to figures (or object surfaces) requires global context information. Border ownership modulation is not a marginal effect. In terms of number of cells affected and strength of modulation, it is comparable to the modulation produced by direction of motion or color. This is understandable since image segmentation and border ownership representation are fundamental requirements for visual object recognition. The representation of local and global information in single cells poses the question of how this information can be read by subsequent stages. We find that orientation and border-ownership selective cells often code color and local contrast polarity as well. The key finding is that various stimulus dimensions are represented factorially. For example, for each orientation, cells exist for all kinds of color selectivity, and for both sides of border ownership. Therefore the single dimensions can be read out by relatively simple mechanisms using linear combinations of the responses of those neurons. In this sense, the various dimensions of visual information are represented explicitly.
This PDF is available to Subscribers Only