Foundations of Visual Perception

1
Foundations of Visual Perception

• Peter Robinson
• School of Physics, University of Sydney
• Brain Dynamics Center, Westmead Hospital
• University of Sydney
• Faculty of Medicine, University of Sydney

2
Overview

• The Visual System.
• The Binding Problem and Gamma Oscillations.
• Brain Modeling.

3
Primary Visual Pathways
Kandel, Schwartz, Jessell (1995)
4
Eye Retina LGN V1
Kandel, Schwartz, Jessell (1995)

• retinal connections highlight edges, etc.
• 100 million cells, only 1 million fibers
• in optic nerve.

• LGN has 6 layers
• 3 ipsilateral (I), 3 contralateral (C).
• 2 magnocellular (movement, depth).
• 4 parvocellular (color, form).

5
Retinotopic mapping

• 1-1 primary mapping from retina to V1.
• Central field (fovea) is disproportionately
  represented.
• Demonstrated by primate experiments.
• Ocular dominance (OD) columns are seen

de Valois de Valois (1990)
6
Ocular Dominance and Orientation Preference

• Orientation preference (OP) varies with
• position in each OD band.
• Singularities, or pinwheels, occur mostly
• near OD band centers.
• V1 is tessellated into hypercolumns.
• Each hypercolumn corresponds to a
• discrete visual field (VF).

Kandel, Schwartz, Jessell (1995)
7
Patchy connections in V1

• Mid-range (few mm) cells in V1 have patchy
  connections.
• Preferentially connect similar feature
  preferences.
• Seen in tree shrew OP data

Bosking, Zhang, Schofield Fitzpatrick (1997)
8
The Binding Problem

• Scenes are broken down and analyzed via many
  pathways, each with
• different feature preference.
• How are these disparate features bound into a
  single percept?
• How are different objects distinguished?

Shadlen et. al. (1999)
Dworetzky (1994)
9
Gamma Oscillations

• Firing of cells in visual cortex can be highly
  correlated.
• Correlation functions (CFs) measure commonality
  of
• firing.
• CFs are highest for nearby cells with similar
  feature
• preference and fall off with separation and
  disparity.
• Do gamma oscillations mediate binding,
• or are they epiphenomena?

Engel, Konig, Kreiter, Schillen, Singer (1992)
10
Modeling

• We use a continuum model at scales of 0.1 mm to
  whole brain
• Cortex is approximated as 2D.
• 1-1corticothalamic mapping.
• Include neural and anatomical properties on
  scales from synapses to whole brain.
• Average over scales below about 0.1 mm.
• Seek equations for resulting neural activity.
• Such models date from 1970s on Nunez, Wilson,
  Cowan, Lopes da Silva, Freeman,
• Wright, Liley, Jirsa, Haken, Sydney group, and
  others.

11
Neurons

• Neural firing underlies all emergent phenomena.
• Excitatory (e) neurons excite others.
• Inhibitory (i) neurons suppress others.
• Inputs via synapses on dendrites.
• Voltage spikes fired when a threshold is
  reached.
• Spikes travel to other neurons via axon
  terminals.
• Cortex contains
• Long-range (several cm) excitatory neurons.
• Mid-range (several mm) excitatory neurons.
• Short-range (lt 1 mm) excitatory neurons.
• Short-range (lt 1 mm) inhibitory neurons.

Kandel, Schwartz, Jessell (2000)
12
Overview of Model Outcomes

• Steady states, linear properties, nonlinear
  dynamics

13
Gamma Correlations

• 1 long bar crossing different VFs produces a
• stronger correlation than 2 separate short bars.
• No correlation for oppositely moving short bars.
• Consistent with summation over subfeatures.
• Consistent with infill of missing contours.

Dworetzky (1994)
Engel, Konig, Kreiter, Schillen, Singer (1992)
14
Scene Segmentation
S1S2

• How do spikes know theyre related.
• Conflicting stimuli presented to 4 sites
• 1 and 2 have vertical OP.
• 3 and 4 have horizontal OP.
• Correlations segment the scene into objects
• Theory explains this effect

S1
S2
S3
Engel, Konig, Singer (2002)
15
Summary

• Each stage in visual processing extracts
  higher-order information, but in multiple
• channels.
• Binding must occur to interrelate these
  features.
• V1 exhibits gamma correlations between cells
  stimulated by related features.
• Physiologically-based brain modeling has been
  verified against multiple experiments.
• Modeling patchy connections explains numerous
  gamma phenomena, including
• frequencies, wavelengths.
• correlation properties.
• scene segmentation.
• gamma waves obey the Schroedinger equation.
• Longer talk ANU Res. School Phys. Sci. Eng.,
  Thu. 20 July.

16
Further Visual Pathways

• Deal with motion, color, depth, form, etc. in
  more detail.
• Link with memory, learning, motor outputs, etc.

Koch (2004)
Koch (2004)