Early Vision

Bruce Draper

Department of Computer Science

Colorado State University

Overview

“The Biomimetic Vision Trilogy”1. Selective Attention

– Understanding the problem– Last week

2. Early Vision– Understanding the literature– Today

3. Ventral vision – Understanding object recognition– March 9 (next week)

General Theme

Vision evolved to serve the needs of animals– Vision is action oriented (it guides behavior)

Actions may be immediate (e.g. grasp, navigate) Actions may be delayed (“perception”)

– Vision is not one system As animals became more complex, more and more visual

capabilities evolved in separate systems

Note: this is not a new idea. See The Visual Brain in Action by Milner & Goodale 1996; The Metaphorical Brain by Arbib 1972; or even Cybernetics by Weiner 1948.

Input: The Eye(s)

Start at the beginning: Lens focuses light Iris serves as aperture Retina contains receptors Optic nerve transmits to brain

Lens, iris are controlled by muscles under the control of the brain

S. Palmer. Vision Science. P. 27

Retina as Processor

• Five cell types:• receptor (rod/cones)

Species dependent• horizontal• bipolar cells• amacrine cells• ganglion cells

• Its inside out!• Blind spot where optic nerve passes through retina

S. Palmer. Vision Science. P. 30

Retina (cont.)

Ganglion Cells– The first cells to produce spike discharges

Other retinal cells use graded potentials Spikes are needed for long distance communication

– On-center off-surround– Off-center on-surround

Two Types of Ganglion Cells

P ganglion cells in primates (like Y cells in cats):– Large receptive fields (low frequency?)– Transient response– Fast transmission– Receive inputs from all colors and from rods

P ganglion cells (like X cells in cats):– Small receptive fields (high frequency?)– Sustained response– Medium transmission– Color opponent channels

Fields of View & Stereo

Right hemisphere receives the left visual field from both eyes

– And vice-versa– Splitting the field of view

supports disparity computations

High resolution in fovea, lower elsewhere

– Fovea is ±2° (thumbnail at arms length)

Visual Projections

The eye + optic nerve is a shared device There are eleven projections (“endpoints”) of the

optic nerve:– Retinogeniculate Projection

Onto LGNd (Lateral Geniculate Nucleus, dorsal) and then to V1 (a.k.a. primary visual cortex, striate cortex)…

This path is dominant in people; barely evident in non-mammals

– Retinotectal Projection Onto Superior Colliculus, then the Pulvinar Nucleus, then LGN,

V1, MT, higher level centers… This path is dominant in non-mammals; evolutionarily older Involved in eye movements, motion, tracking

Projections (LGN & S.C.)

S. Palmer. Vision Science. P. 25.

Projections (II)

At least 8 more (minor) projections!– Retina Suprachiasmatic Nucleus

Circadian rhythms SN is part of the hypothalamus (like LGNd) SN receives multi-modal projections

– Retina Nucleus of the Optic Tract (NOT) Optokinetic nystagmus

– Retina Accessory Optic Nuclei Visual control of posture, locomotion

– Retina Pretectum Pupillary Light Reflex

Two LGNd Channels

P cells in the retina project to the two magnocellular (“large cell”) layers in the LGN.

– Livingstone & Hubel: color-blind, fast, high contrast sensitivity, low spatial resolution

P cells in the retina project to the four parvocellular (“small cell”) layers in the LGN.

– L&H: color selective, slow, low contrast sensitivity, high spatial resolution

LGNd also has interlaminar layers with unknown role/properties

– Receives projections from optic nerve, S.C.

Right at the levels of cells; wrong at the level of populations

Primary Visual Cortex (V1)

First cortical visual area– Columnar (like all cortex)

Retinotopically mapped Ocular dominance columns Edges (Gabor filters), color,

disparity & motion maps Connects to other

retinotopic areas (V2, V3, MT)

http://webvision.med.utah.edu/imageswv/capas-cortex.jpg

Proof of Retinotopic Mapping

Pattern flashed (like a strobe) in front of monkey

injected with sugar dye

Left primary visual cortex of the same

monkey

Tootell, et al. 1982.

Single Cell Recordings in V1

Two Channels in V1

P Magnocellular LGN Layers layer 4C of V1.– Projects further to layer 4B

Motion direction selective Orientation selectivity & binocular No color

P Parvocellular LGN 4C of V1– Projects to layers 2 & 3.

Layers 2 & 3 subdivide into “blobs” and “interblobs”:– Blobs are color selective, simple receptive fields– No orientation/movement direction selectivity, binocularity.– Prefer low frequencies, have small Magnocellular input – Interblobs have reduced (non-zero) color selectivity– Binocular, high-frequency, orientation selective

Simple Cells

The first orientation selective cells found in V1 were labeled “simple cells” – Well approximated by Gabor functions with fixed

orientations, scales and phases

Jones & Palmer 1987

Complex Cells

The second set of orientation selective cells were called complex cells– Well modeled as combining 90° out-of-phase

Gabor responses (quadrature pairs)– Captures energy at a particular orientation &

scaleevenfilter

oddfilter

+

Organization of cells in V1

Hubel & Weisel proposed the following organization for cells in V1

Well, a little more complex…

Ocular dominancecolumns

Color-coded orientationSensitivity columns

More on V1

Only about 27% of V1 cells are orientation selective About 70% of orientation selective cells are complex

cells Orientation selective cells also include end-stopped

cells (a.k.a. hypercomplex cells), and grating cells. Non-orientation selective cells include cells that

respond to:– Color (Hue/Sat maps?)– Disparity– Motion

V1 Connections

V1 is the starting point of cortical visual processing.

Dorsal projections lead to somatosensory and motor control areas

Ventral projections lead toward associative memories

From Van Essen 1992. Image can be found athttp://webvision.med.utah.edu/imageswv/Visual-Cortex1.jpg

Anatomical Maps of Visual Cortex

1983 Version 1990 Version

Two Visual Subsystems

In 1982, Ungeleider & Mishkin propose that there are two primary visual pathways in humans and primates:– The dorsal (or “what”) pathway

Ends in the posterior parietal cortex

– The ventral (or “where”) pathway Ends in the inferotemporal cortex

Visualizing Two Subsystems

D. Milner & M. Goodale, The Visual Brain in Action, p. 22