A Neurophysiology Model that Makes Quantifiable Predictions of Geometric Visual Illusions

Author: Nasim Nematzadeh

Nematzadeh, Nasim, 2018 A Neurophysiology Model that Makes Quantifiable Predictions of Geometric Visual Illusions, Flinders University, College of Science and Engineering

Terms of Use: This electronic version is (or will be) made publicly available by Flinders University in accordance with its open access policy for student theses. Copyright in this thesis remains with the author. You may use this material for uses permitted under the Copyright Act 1968. If you are the owner of any included third party copyright material and/or you believe that any material has been made available without permission of the copyright owner please contact copyright@flinders.edu.au with the details.

Abstract

This thesis is focused on modelling low-level vision and the encoding of visual input by the

simple retinal cells.

The study of human vision is a ‘multidisciplinary field’, connecting the physiology of vision to

bioplausible computational modelling, as well as psychophysical experiments with real subjects. One

source of evidence about vision is optical illusions, which do not necessarily occur in a computer

vision (CV) model, but should be apparent in a vision model that claims to represent the way human

vision works, or a vision system that tries to identify the same patterns and features that a human

would. This area of research leads to a shibboleth for testing bioplausible models of vision. The goal

of this PhD research is to describe, simulate and quantify a bioplausible model that reflects

differences in the dominant tilts apparent in a family of Geometrical Illusions.

In this dissertation, a neurophysiologically inspired model has been developed, implementing

the lateral inhibition of the retinal cells based on Gaussians (Mexican Hat) filtering at multiple scales.

Our model produces Difference of Gaussian (DoG) at different scales as a bioplausible

representation of the image and interprets them as edge maps at multiple scales. The edge map is

further investigated using an analytic processing pipeline in Hough space to quantify the angle of tilt

emergent in the model around four reference orientations (-45°, 0°, 45°, 90°). In this study, a

quantifiable prediction is developed for the degree of perceived tilt in the Café Wall pattern, a typical

Geometric Illusion, in which the mortar between the tiles seems to converge and diverge. The model

also predicts different perceived tilts in different areas of the fovea and periphery as the eye saccades

to different parts of the image. Several sampling sizes and aspect ratios, modelling variant foveal

views, are investigated across multiple scales in order to provide confidence intervals around the

predicted tilts, and to contrast local tilt detection with a global average across the whole Café Wall

image. Beyond the Café Wall illusion, the model has been applied to investigate local tilt for a more

general class of complex Tile illusions such as Complex Bulge pattern and Spiral Café Wall.

This is the first model to provide verifiable quantitative predictions of the tilt perceived across

a range of “Café Wall” illusions. More formally, we have shown that a simple Difference of Gaussian

Classical Receptive Field model, implementing multiscale responses of a symmetrical ON-center

and OFF-surround Retinal Ganglion Cells (RGCs), can easily reveal the emergence of tilt in these

patterns. We hypothesize that in later stages of perception, these local tilt cues are integrated by

more complex cortical cells.

Keywords: Visual perception, Cognitive system, Pattern recognition, Biological neural network, Self-organising systems, Classical Receptive Field (CRF) models, Geometrical illusions, Tilt effect, Difference of Gaussians, Perceptual grouping, Gestalt grouping principles

Subject:

Thesis type: Doctor of Philosophy
Completed: 2018
School: College of Science and Engineering
Supervisor: Prof. David Powers