2016/12/23

Lecture 6

The Seeing Brain:

Constructing Our Visual World

田意民 老師
tien@csmu.edu.tw

Reading
• 本主題涵蓋課本第六章內容
– Chapter 6: The seeing brain

1
 2016/12/23

Contents
• From Eye to Brain
• Cortical Blindness and “Blindsight”
• Functional Specialization of the Visual Cortex
Beyond V1
• Recognizing Objects
• Category Specialization in Visual Object
Recognition?
• Recognizing Faces
• Vision Imagined

Contents
• From Eye to Brain
• Cortical Blindness and “Blindsight”
• Functional Specialization of the Visual Cortex
Beyond V1
• Recognizing Objects
• Category Specialization in Visual Object
Recognition?
• Recognizing Faces
• Vision Imagined

2
 2016/12/23

Seeing as a Constructive
Process

The Kanizsa illusion

From Eye to Brain

「右腦管左眼，左腦管右眼」這是錯的！
人類是視野控制！

3
 2016/12/23

From Eye to Brain

• Note: left space to right brain (not left eye
to right brain)

From Zeki (1993).

Copyright ©
Blackwell
Publishing.
Reproduced with
permission.

Geniculostriate Pathway
• Number of different pathways from eye to
brain
• Main route terminates in primary visual
cortex (V1)
• Route called geniculostriate pathway
because it goes via lateral geniculate
nucleus (LGN) and terminates in striate
cortex (another name for V1)

4
 2016/12/23

Lateral Geniculate Nucleus

網膜與神經的反應和螢幕的圖點不同

Lateral Geniculate Nucleus

• Contains six layers,
three for each eye
• Cells have a center-
surround receptive field
• They respond to
differences in light
across their receptive
field (e.g. presence of
light in center, absence
in surround)

5
 2016/12/23

Primary Visual Cortex (V1)

邊界、走向、波長
• Extracts basic information from the visual scene
(e.g. edges, orientations, wavelength of light)
• This information is used by later stages of
processing to extract information about shape,
colour, movement, etc.
• Single-cell recordings by Hubel and Wiesel lead
to a hierarchical view of vision in which simple
visual features (e.g. points of light) are combined
into more complex ones (e.g. adjacent points of
light may combine into a line)
點、線、邊界、長度、平面

Cells of Primary Visual Cortex (V1)

• Simple cells may derive their response by combining the
responses of several LGN centre-surround cells
• Simple cells respond to different orientations

LGN V1
From Zeki (1993). Copyright © Blackwell
Publishing. Reproduced with permission.

6
 2016/12/23

Cells of Primary Visual Cortex (V1)

• Complex cells may be derived by combining the
responses of several simple cells
• Complex cells respond to orientation too but
have larger receptive fields and require
stimulation on their entire length (simple cells
also respond to points of light)
• Hypercomplex cells (outside V1) may be derived
by combining the responses of several complex
cells
• Unlike complex cells, they are sensitive to length
as well as orientation
點、線、邊界、長度、平面

Contents
• From Eye to Brain
• Cortical Blindness and “Blindsight”
• Functional Specialization of the Visual Cortex
Beyond V1
• Recognizing Objects
• Category Specialization in Visual Object
Recognition?
• Recognizing Faces
• Vision Imagined

7
 2016/12/23

Spatial Arrangement of Primary Visual

Cortex (V1)
• Retinotopic organization –
the spatial arrangement of
light on the retina is retained
in the response properties of
V1 neurons (except inverted)
• Damage to parts of area V1
results in blindness for the
corresponding region of
space (e.g. hemianopia半側
盲)

Adapted from Zeki (1993).

Cortical and Sub-cortical Vision

• Damage to
geniculo-
striate route
impairs
conscious
vision, but
other aspects
of vision
spared
(blindsight)

8
 2016/12/23

Blindsight盲視
• Damage to V1 leads to a clinical diagnosis of
blindness (the patient cannot consciously report
objects presented in this region of space)
• However, the patient is still able to make some
visual discriminations in the "blind" area (e.g.
orientation, movement direction) – called
blindsight
• This is because there are other routes from the
eye to the brain
• The geniculostriate route may be specialized for
conscious vision but other routes act
unconsciously

Blindsight (cont.)
• Filling-in(填補) of ‘blind’ regions similar to
filling-in of normal blind spot

Adapted from Torjussen (1976).

9
 2016/12/23

Contents
• From Eye to Brain
• Cortical Blindness and “Blindsight”
• Functional Specialization of the Visual Cortex
Beyond V1
• Recognizing Objects
• Category Specialization in Visual Object
Recognition?
• Recognizing Faces
• Vision Imagined

Beyond Visual Cortex

• Visual cortex (striate and extrastriate)

extracts basic visual information – colours,
movement, shapes, edges
• In order for this information to be used it
needs to make contact with other types of
information…
– Where the object is in space (and this can’t be
computed from the retinal image alone)
– What the object is

10
 2016/12/23

Beyond V1
Where

Broader receptive field,

Less coherently organized in space,
Divide and conquer What

Area V4 and Area V5/MT

11
 2016/12/23

Area V4 and Area V5/MT

• PET study: Zeki et al. (1991)

• Coloured images
("Mondrians") compared to
greyscale equivalents
 area V4 active, conclude this
region is specialized for colour
• Moving dots compared to
static dots
 area V5/MT active, conclude
this region is specialized for
visual movement

Colour Perception and Area V4

• Why does the brain need a specialized colour

centre given that the retina is sensitive to
different wavelengths of light?
• The problem is that wavelength depends on the
composition of the light source (e.g. daylight,
electric light) as well as the colour of an object
• Area V4 tries to compute the colour of the object
taking into account variations in lighting
conditions
• This is called colour constancy

12
 2016/12/23

Colour Perception and Area V4

• Cells in V4 continue to respond to the same surface
colour if the light source is changed, whereas cells in V1
do not
• Patients with damage to area V4 see the world in black
and white – they are called achromatopsic (not to be
confused with colour blindness due to cone deficiency)
• Although achromatopsic patients fail to see colour, their
retina and their V1 cells still respond to different
wavelengths of light
• This is another example of how visual perception is
constructed, rather than the mere detection of physical
properties in the environment

Movement Perception and Area

V5/MT
• Cells in V5/MT do not respond to colour
but 90% of them respond to particular
directions of movement
• Patients with bilateral damage to this
region see the world in a series of still
frames
• They are said to be akinetopsic(無動視覺)
• The patients can detect movement in other
senses (e.g. hearing, touch)

13
 2016/12/23

Biological Motion

BM的控制可能比一運動知覺更加特化
• It is possible to discriminate biological from random
motion given an array of moving dots
• Brain imaging and neuropsychology suggest that this
may use different regions/mechanisms to determining
the overall direction of movement
• Akinetopsic patients can discriminate biological motion

Visual illusions
• Illusory objects
and illusory
motion activate
same parts of
brain as real
vision

http://www.michaelbach.de/ot/#history

Top image by Isia Levant, 1981,

www.michaelbach. de/ot/mot_
enigma/index.html

14
 2016/12/23

Contents
• From Eye to Brain
• Cortical Blindness and “Blindsight”
• Functional Specialization of the Visual Cortex
Beyond V1
• Recognizing Objects
• Category Specialization in Visual Object
Recognition?
• Recognizing Faces
• Vision Imagined

A Model of Object Recognition

WHAT

From Riddoch and Humphreys (2001).

15
 2016/12/23

A Model of Object Recognition

• Four broad stages…
1. Early visual processing (colour, motion,
edges, etc.)
2. Grouping of visual elements (Gestalt
principles, figure–ground segmentation)
3. Matching grouped visual description onto a
representation of the object stored in the
brain (called structural descriptions)
4. Attaching meaning to the object (retrieved
from semantic memory)

Combining Parts into Wholes:

Gestalt Grouping
• Constitutes the second proximity similarity
stage of the model of
object recognition
– Law of proximity
– law of similarity
– law of good continuation
– law of closure
good continuation closure
– law of common fate (e,
not shown)

16
 2016/12/23

Seeing Parts but not Wholes:

Integrative Agnosia

• Disorders in object recognition are called agnosia

• Many different types of agnosia that can broadly be
divided into
– disorders of perception (apperceptive agnosia) or
– disorders of meaning (associative agnosias)
• Integrative agnosia is a type of apperceptive agnosia in
which grouping principles are disrupted
• This prevents stored knowledge of objects being
accessed, but does not prevent the patient from seeing
basic visual elements (computed in stage 1)

Seeing Parts But Not Wholes:

Integrative Agnosia (cont.)

Adapted from Humphreys and Riddoch (1987)

and Riddoch and Humphreys (1995)

17
 2016/12/23

Seeing Parts but not Wholes:

Integrative Agnosia

See p.116. fig.

Routes to Object Constancy

• Object constancy achieved by mapping a potentially
infinite number of visual depictions on to a finite set of
stored descriptions of the structure of objects
• H1: One suggestion is that the brain stores objects in a
single viewpoint (the canonical viewpoint that contains
the principal axis)
• In this account, object recognition involves view
normalization from the seen viewpoint to the stored
viewpoint (mental rotation)
• H2: Another suggestion is that stored structural
descriptions are accessed by matching feature-by-
feature

18
 2016/12/23

Object Constancy via View

Normalization?
• Patients with parietal lobe damage (on right) may be
unable to recognize objects in unusual views but able to
recognize them in canonical views
• Other patients can recognize objects in all viewpoints
(unusual and canonical) but cannot choose the correct
orientation for an object
• This is called object orientation agnosia and it provides
evidence that the principal axis is stored separately from
other aspects of object recognition
See p.118. fig.

Routes to Object Constancy (cont.)

• This neuropsychological test examines object constancy
• Does recognizing objects from unusual views involve
mental rotation to a standard view?

From Riddoch and Humphreys (1995).

19
 2016/12/23

Object Constancy via View

Normalization?

Patient EL

From Harris et al. (2001). Copyright © The

MIT Press. Reproduced with permission.

Neural Substrates of Object Constancy

• Monkey cells in IT (inferotemporal) cortex respond to

very particular object attributes (e.g. corners, shapes)
but are less concerned with where they are located in
space (Gross, 1992)
• These are ideal conditions for computing object
constancy
• fMRI in humans shows that inferotemporal regions
respond to the same object presented in different sizes –
left region is insensitive to viewpoint but right region is
viewpoint sensitive
• This is consistent with 2 different routes to object
constancy(其實只是兩個笨系統整合而已)

20
 2016/12/23

"What" versus "How" and "Where"

• Object constancy is primarily concerned with identifying

what an object is as opposed to where it is
• The pathway from V1 to inferotemporal cortex is termed
the "what" or ventral route

• Another pathway exists from V1 that terminates in the

parietal lobes
• This route is involved in locating objects in space and
interacting with them and has been called the dorsal
route, the "how" route and the "where" route

"What" versus "How" and "Where"

21
 2016/12/23

Contents
• From Eye to Brain
• Cortical Blindness and “Blindsight”
• Functional Specialization of the Visual Cortex
Beyond V1
• Recognizing Objects
• Category Specialization in Visual Object
Recognition?
• Recognizing Faces
• Vision Imagined

Category Specificity
• Domain-specific modules (Fodor, 1983)

• Object, Face, Word

22
 2016/12/23

Contents
• From Eye to Brain
• Cortical Blindness and “Blindsight”
• Functional Specialization of the Visual Cortex
Beyond V1
• Recognizing Objects
• Category Specialization in Visual Object
Recognition?
• Recognizing Faces
• Vision Imagined

The Problem with Faces

 [Humpty Dumpty to Alice]
"You’re so like other
people…the two eyes
(marking their place in the
air with his thumb), nose
in the middle, mouth
under. It’s always the
same. Now if you had
the two eyes on the same
side of the nose, for
instance – or the mouth
at the top – that would be
some help"
Bettmann/CORBIS

23
 2016/12/23

The Problem with Faces

• Face recognition is a within-category

discrimination (all faces look the same), whereas
other object recognition is between category (e.g.
distinguishing a pen from a cup)
• Maybe faces require different types of
processing to other objects?
• Maybe faces are so important from a
social/evolutionary perspective that they have a
mechanism all to themselves? = domain-
specificity

Different Aspects of Face Processing:

Bruce and Young (1986)

From Parkin (1996).

24
 2016/12/23

Evidence for Bruce and Young (1986) Model

• Double dissociation between recognizing familiar faces
and matching unfamiliar faces across different viewing
conditions (face constancy)
• In face naming, it is often possible to retrieve semantic
facts without retrieving the name but the reverse pattern
is not found (i.e. name generation depends on semantic
retrieval)
• Double dissociation between recognizing familiar faces
and recognizing emotion, age and sex
• Double dissociation between recognizing familiar faces
and using lip-reading cues

Are Faces Special? Prosopagnosia

• Prosopagnosia = impairments of face

processing that do not reflect difficulties in early
visual analysis
• (also used specifically to refer to difficulty in
recognizing previously familiar faces)
• De Renzi (1986) – patient failed to recognize his
own family but could do so by voice, clothes “are
you…? [wife’s name] I guess you are my wife
because there are no other women at home, but
I want to be reassured”
• Could match different views of faces and name
other objects

25
 2016/12/23

Haxby et al. (2000)

Are Faces Special? Fusiform Face Area

• Responds to faces
more than other types
of objects in
functional imaging L R
experiments
(Kanwisher)
• But this may be a
relative difference
between faces and
objects rather than an R L
absolute difference

26
 2016/12/23

Why Are Faces Special?

(1) Task difficulty?

(2) Holistic/configural processing?
(3) Visual expertise?
(4) Domain-specificity?

Task Difficulty?

• Farah et al. (1995) devised a task involving

faces and spectacles that was equally difficult for
controls (both 85% correct)
• Prosopagnosic patient, LH, was normal on
spectacles (92%) but impaired on faces (62%)
• Other patients are reported who are bad at
recognizing familiar objects but not faces (so
faces are not always hard)

27
 2016/12/23

Holistic Processing Disrupted by

Inversion

Holistic Processing Disrupted by

Inversion

28
 2016/12/23

Holistic Processing Disrupted by

Inversion

From Thompson, P. (1980). Copyright ©

Pion Limited, London. Reproduced with
permission.
© Pion Limited

Holistic Processing Disrupted by

Inversion (cont.)

From Thompson, P. (1980). Copyright ©

Pion Limited, London. Reproduced with
permission.
© Pion Limited

29
 2016/12/23

Holistic/Configural Processing?

• Farah (1990) – all visual

recognition lies on a
continuum between
recognition by parts and
recognition by wholes
• Faces may lie at one end
of the continuum (holistic),
words may lie at the other
end (part-based) and
objects are intermediate
• She presents evidence
that face and word
recognition impairments
always affect object
recognition

Holistic/Configural Processing?

• Subsequent
researchers have
argued that faces,
visual words, and
objects can be
independently
impaired
• This may suggest
separate stores rather
than a single
continuum

30
 2016/12/23

Visual Expertise?
Images provided courtesy
of Michael J. Tarr (Carnegie
Mellon University,
Pittsburgh), see
www.tarrlab.org

• Gauthier and colleagues suggest that faces are special

because we have become experts at within-category
discriminations
• Claims that becoming an expert at "Greeble"
discrimination involves the fusiform face area, as do
other types of within-category discrimination (e.g. model
car collectors)

Visual Expertise?

• BUT: do Greebles look like faces?

• BUT: not all prosopagnosic patients
impaired on within-category discrimination
• Patient WJ – owned a flock of sheep and
could distinguish between them
• Patient RM – could distinguish between
his collection of 5000 miniature cars

31
 2016/12/23

Visual Imagery
• Imagery =
perception in
reverse?
• But how low does it
go? Kosslyn
argues V1 is
important but ‘how
low it goes’ may
depend on the
content of vision
(e.g. whether it is
faces, colors or
lines)

Domain-specificity?

• It is the default hypothesis when all else fails

• Strongest evidence for this comes from
prosopagnosic patients who can make within-
category discriminations
• Even in this instance, it is debatable that the
tasks are matched (e.g. 36 sheep versus huge
numbers of faces)
• Problem for proving this hypothesis is that it is
hard to find stimuli that are just like faces (in
terms of processing, difficulty, etc.) but that are
not faces

32
 2016/12/23

Summary and key points

• The properties of V1
• Specialized functions of visual areas beyond V1
• Object recognition and categorical recognition of
objects
• Why face recognition is special?
• Mental imagery vs. perception

Reading
• 本主題涵蓋課本第六章內容
– Chapter 6: The seeing brain

• Further reading: p.127

33
 2016/12/23

Next Topic:
The spatial brain: Attention
• Please read following chapters before next
lecture
– Chapter 7: The spatial brain

34