Advanced

Occluding Contours: A Computational Model of Suppressive Mechanisms in Human Contour Perception

Månsson, Jens LU (2000)
Abstract
A fundamental problem in vision is how to identify the occluding contours of objects and surfaces, given the ambiguity inherent in low-level visual input. A computational model is proposed for how occluding contours could be identified by making use of simple heuristics that reduce the ambiguity of individual features. In the striate cortex, a large majority of cells are selective for both contrast and orientation; i.e., they respond preferentially to simple features like contrast edges or lines. The heuristics we propose enhance or suppress the outputs of model striate-cortical cells, depending on the orientation and spatial distribution of stimuli present outside of the "classical" receptive field of these cells. In particular, the... (More)
A fundamental problem in vision is how to identify the occluding contours of objects and surfaces, given the ambiguity inherent in low-level visual input. A computational model is proposed for how occluding contours could be identified by making use of simple heuristics that reduce the ambiguity of individual features. In the striate cortex, a large majority of cells are selective for both contrast and orientation; i.e., they respond preferentially to simple features like contrast edges or lines. The heuristics we propose enhance or suppress the outputs of model striate-cortical cells, depending on the orientation and spatial distribution of stimuli present outside of the "classical" receptive field of these cells. In particular, the output of a cell is suppressed if the cell responds to a feature embedded in a texture, in which the "component features" are oriented in accordance with the orientation-selectivity of the cell. The model has been implemented and tested on natural as well as artificial grey-scale images. The model produces results that in several aspects are consistent with human contour/form perception. For example, it reproduces a number of known visual phenomena such as illusory contours, contour masking, pre-attentive pop-out (due to orientation-contrast), and it enhances contours that human observers often report perceiving as more salient. (Less)
Please use this url to cite or link to this publication:
author
organization
publishing date
type
Working Paper
publication status
published
subject
keywords
Cognitive Studies
language
English
LU publication?
yes
id
0c41bee7-e8e2-4c90-b48a-31df6e59807c (old id 525876)
alternative location
http://www.lucs.lu.se/Abstracts/LUCS_Studies/LUCS81.html
date added to LUP
2007-10-08 16:19:29
date last changed
2016-04-16 11:32:48
@misc{0c41bee7-e8e2-4c90-b48a-31df6e59807c,
  abstract     = {A fundamental problem in vision is how to identify the occluding contours of objects and surfaces, given the ambiguity inherent in low-level visual input. A computational model is proposed for how occluding contours could be identified by making use of simple heuristics that reduce the ambiguity of individual features. In the striate cortex, a large majority of cells are selective for both contrast and orientation; i.e., they respond preferentially to simple features like contrast edges or lines. The heuristics we propose enhance or suppress the outputs of model striate-cortical cells, depending on the orientation and spatial distribution of stimuli present outside of the "classical" receptive field of these cells. In particular, the output of a cell is suppressed if the cell responds to a feature embedded in a texture, in which the "component features" are oriented in accordance with the orientation-selectivity of the cell. The model has been implemented and tested on natural as well as artificial grey-scale images. The model produces results that in several aspects are consistent with human contour/form perception. For example, it reproduces a number of known visual phenomena such as illusory contours, contour masking, pre-attentive pop-out (due to orientation-contrast), and it enhances contours that human observers often report perceiving as more salient.},
  author       = {Månsson, Jens},
  keyword      = {Cognitive Studies},
  language     = {eng},
  title        = {Occluding Contours: A Computational Model of Suppressive Mechanisms in Human Contour Perception},
  year         = {2000},
}