Visual Perception

1
Visual Perception
1. Memory tests were run 2. Perception of simple
and complex forms (masking)
memory works a lot better if we know what we are
supposed to be looking at
perception is a filtering mechanism
2
Masking
Experiments with a tachistoscope have shown that
there is a difference in perception time between
simple forms (figures in Gestalt terminology) and
more complex ones. The letters A, U and T are
simple forms and on an uncluttered white
background (field) an exposure time of 100 msec
is easily sufficient to perceive them as these
forms.
U
However, if the same letters are exposed for the
same length of time within a ring then 100 msec
is not sufficient to construct this complex form
and only the ring is perceived.
The same is true if the masking ring is
presented no more than 50 msec BEFORE or AFTER
the exposure of the letter. Between 50 and 100
msec delay the letter may be perceived. Delays
of longer than 100 msec result in the ringletter
form being seen.
3
Theories of Visual Perception by Ian E.
Gordon published by John Wiley Sons,
1989. Theories must meet certain criteria
Should offer economical accounts of a range of
facts. A theory is not much use if a description
of it is a long as that required to describe the
relevant phenomena
Should attempt to explain phenomena, or at least
suggest causal links between them
Should be testable should be stated in such a
way that deductions can be derived and tested
empirically
4
There is considerable variation in the style and
language of theories of visual perception
The reason for this is essentially that none of
them deals with exactly the same arena of
perception
The greater the number of regions to be included
in a theory the more that theory tends to be
general in form
5
ENVIRONMENT
ENVIRONMENT
receptors
Brain
stimuli
effectors
ENVIRONMENT
ENVIRONMENT
6
The environment is the physical world of surfaces
and objects - the ecology of the organism
Incoming stimuli from objects in the world give
rise to events some of which can be detected by
perceivers
Knowledge of the important properties of stimuli
has come mainly from physics
Sensory surfaces take incoming stimuli and
translate them into a neural code it is
important to know the nature of this
transduction, how light is absorbed by the eye
7
Important questions concern the pathways taken by
neural messages, the codes which are used to
represent differences in quality, intensity and
duration
The brain has an obvious role play
Most behaviour depends upon brain processes but
these are commonly not available to direct study
and must be explored indirectly
Organisms make explicit responses to stimuli in
the environment
8
Responses
The pupil constricts in response to light and
sweat is produced by very brief exposure to taboo
words
This can be used as a sign that the words have
been detected and that they have induced
emotional responses
The quickest action any human is capable of is an
eye movement
we move around in the world and in this way
partially determined to stimulation we receive
It has been discovered that the eye takes in much
less information during an eye movement then when
it is stationary
We may make eye movements which are abrupt and
ballistic and also movements which are smoothly
graded
what guides the selection of appropriate
movements?
9
therefore
Regions of interest to perceptual theorists
THE ENVIRONMENT
INCOMING STIMULATION
RECEPTOR SURFACES AND THE PERIPHERAL SENSORY
NERVOUS SYSTEM
THE BRAIN
PERIPHERAL EFFECTOR PROCESSES
MOTOR RESPONSES BY THE PERCEIVER
10
GESTALT

• Wertheimer (1880-1943)
• Koffka (1886-1941)
• Köhler (1887-1964)

"why do things look as they do?"
Their statement of intent
what must be explained by perceptual theories is
the stability and coherence of the world of
everyday experience
Pencil and hole in page and your nose
11
GESTALT
Whenever we open our eyes we see objects and
surfaces, not sensations of light.
We can easily distinguish between figure and
ground (the figure possesses Gestaltqualität) -
ground is less distinct.
12
GESTALT
distinguishing between figure and ground
figure
ground
13
GESTALT
distinguishing between figure and ground
But is not usually a problem in a
three-dimensional world
Differentiation between figure and ground can be
confused in a two- dimensional image, such as
that on a page or a screen
A white circle or a hole in the black triangle?
14
GESTALT
there is a general, underlying principle behind
the numerous examples of organisation which they
discovered (see next few slides)
Gestalt theorists also laid much emphasis on the
simple idea that the whole is greater than the
sum of the parts
See also Peter Checklands book Systems
Thinking Systems Practice
The Muller-Lyer illusion the individual lines
are objectively the same but their relationship
with the arrows creates an illusion which could
not be predicted from knowledge of the individual
components
15
GESTALT
Gestalt theorists also laid much emphasis on the
simple idea that the whole is greater than the
sum of the parts
The Muller-Lyer illusion
16
GESTALT
natural organisation
17
grouping by columns
grouping by rows
equal proximity no dominant direction of grouping
18
grouping by similarity
grouping by continuation
19
GESTALT
The Gestalt movement took a phenomenological
approach rather than an introspective approach to
perception.
Their explanation of perceptual and related
phenomena took the form of hypothetical brain
processes
advanced the nativist theory
The philosophy of Emmanuel KANT was important to
Gestalt Theorists
these were part of a psycho-neural isomorphism
this is inherently nativist in its implications
concerning the origins of perception in the
individual perceiver
20
Empiricists
Richard Gregory born 1923
concluded that perceiving is an activity
resembling hypothesis formation and testing.
Signals received by the sensory receptors trigger
neural events
Appropriate knowledge interacts with these inputs
to create psychological data.
On the basis of such data, hypotheses are
advanced to predict and makes sense of events in
the world
This chain of events is the process we call
perceiving
21
Empiricists
The main arguments are
perception allows behaviour to be generally
appropriate even to non-sense object
characteristics
perception can mediate zero-time delay reactions
when we see only 3 legs on a table
perception can be ambiguous if a single physical
pattern can induce 2 different percepts, then
perception cannot be tied to stimulation in a
one-to-one manner
Neckar cube
22
The NECKAR CUBE
23
Empiricists
The main arguments are
perception can extract familiar objects from a
cluttered background
perception seems to be aided by knowledge
thrushes searching image
atoo talur
but stereotyped, well reinforced knowledge can
refute actual perception so that even if we know
a hollow mask is hollow we still perceive it as a
normal face
Even impossible designs are rationalised
perception can be paradoxical
Penrose design
24
The PENROSE DESIGN
25
Brunswik - Reiter Faces
Brunswik started the idea that perception
involved probabilitistic functionalism
He arrived at this conclusion after a series
of experiments involving perceptions of a
minimal set of facial characteristics
26
Brunswik-Reiter schematic faces
Variation in
nose length
forehead height
mouth height
eye separation
27
Brunswik-Reiter schematic faces
Experimental design
categorise according to these scales
gay-sad young-old good-bad likeable-unlikeable bea
utiful-ugly intelligent-unintelligent energetic-un
energetic
associations found between
apparent mood and age character, likeability and
beauty intelligence and energy
28
Good
Happy
Bad
Sad
29
Brunswik-Reiter schematic faces
Higher the mouth gayer younger the face
but the lower the apparent intelligence
longer noses generally had unfavourable effects
High foreheads received favourable judgements
30
Brunswik-Reiter schematic faces
Conclusion
strong impressions can be induced by very simple
patterns
small changes can induce marked changes in the
impressions they induce
these impressions are not particularly culturally
biased - they are human stereotypes
31
Brunswik-Reiter schematic faces
Conclusion
There is a large amount of constancy, stability
in an inherently unstable world similar to De
Bonos assertions - anything that looks like the
pattern is presumed to be the pattern and we act
accordingly without further thought being
involved
the ecology of the perceiver is important
small changes are important - changes are what
stimulate active recognition
32
David MARR - the computational theory of
perception
worked at Cambridge University
His work on artificial intelligence led to
numerous papers on perception and finally to his
book Vision which was published posthumously (he
died age 35 in 1980)

• important developments which contributed
• to his theories
• information theory
• cybernetics
• construction of large digital computers

33
M A R R
contributing studies used in developing his theory
MARR
cells respond differentially to lines and edges
according to the orientation of these stimuli
cat visual cortex
this suggests that the visual system analyses
visual inputs into specific components, and that
the mechanisms which do this are "wired into" the
nervous system
M A R R
it is therefore possible that the perception of
certain basic features of the world is unlearned
34
M A R R
contributing studies used in developing his theory
MARR
Random dot stereograms
a powerful illusion of depth arises because the
paired stereograms contain central portions which
differ slightly
this triggers normal stereopsis disparity of
left and right views
the strange and wonderful thing about the random
dot stereograms is that the disparity is not
visible - the arrays contain no hint of form
M A R R
this proves that the visual system can extract
disparity information in the absence of pattern
recognition
35
M A R R
contributing studies used in developing his theory
MARR
spatial frequencies of test gratings
staring for a time at a particular grating
reduces sensitivity to that grating temporarily
displays containing black and white stripes of
varying number of stripes per degree of visual
angle
this is not a general loss of visual acuity
because sensitivity to other spatial frequencies
remains unchanged
in the cat visual cortex cells are differentially
sensitive to particular spatial frequencies. It
would appear that one could consider the acuity
of vertebrate visual systems in terms of tuned
channels
M A R R
36

• computation theory
• algorithm
• hardware

M A R R
MARR
Vision must start with the image on the retina
the end point is our awareness of the world
Marr argues that symbolic representations of
various aspects of the world, initially obtained
from the retinal image, are combined into the
descriptions which we call seeing.
There seems to be a picture of the world
available to us whenever we open our eyes and
look around. But the fact is that light stops at
the retina. There can be no actual pictures in
our heads, only neural activity. It follows that
this neural activity is representing the world
symbolically, and we must therefore strive to
understand this symbolic process.
M A R R
37

• computation theory
• algorithm
• hardware

M A R R
MARR
perception of contours
why is it important to be able to perceive
contours - what use is this to the perceiver? In
other words, what of importance in the real world
correlates with contours in the visual image?
Why should the visual system work to make them
explicit?
How might contours be represented symbolically in
our heads?
it is likely to arise as an edge - a feature
which reveals discontinuities between the
surfaces of different objects
M A R R
a n s w e r
matches Brunswik's concern over the ecological
validity of cues
38

• computation theory
• algorithm
• hardware

M A R R
MARR
perception of contours
quite a lot is known about contour perception
the information passed on is about rates
of change, not homogenous illumination or grading
of intensity, which is "extracted" as relevant
contour information

• computation theory

excitatory and inhibitory fields in the retina,
which respond to entire edges, would be needed
for such a mechanism, with rules for interactions
of fields

• algorithm

M A R R
hardware needed for such an algorithm is present
in the retinal ganglion cells which act through
mutual inhibition
a theory in which the main job of vision is to
derive a representation of shape

• hardware

39

• computation theory
• algorithm
• hardware

M A R R
MARR
He also used his knowledge of computer science to
formulate a guiding principle - modular design
in solving computational problems generally, it
is wise to break down the computation into
component parts which should proceed as
independently as possible

• image
• primal sketch
• 2½-D sketch
• 3-D model

M A R R
40
M A R R

• image
• primal sketch
• 2½-D sketch
• 3-D model

MARR
image the retinal image is a spatial
distribution of intensity values
primal sketch takes raw intensity values of the
image and makes explicit certain forms of
information contained therein. The most
important information concerns the spatial
distribution of intensity changes and how they
are organised- allows the possible detection of
surfaces
M A R R
41
M A R R

• image
• primal sketch
• 2½-D sketch
• 3-D model

MARR
2½-D sketch orientation and rough depth of
visible surfaces are made explicit it is as if a
"picture" of the world is beginning to emerge.
However, what is emerging is organised with
reference only to the viewer, it is not yet
linked to a stable, external environment
3-D model representation the shapes and their
orientation become explicit as tokens of
three-dimensional objects organised in an
object-centred framework i.e. independent of
particular positions and orientations on the
retina. By this final stage of vision the
perceiver has attained a model of the external
world
M A R R
42

• primal sketch

M A R R
MARR
from the array of intensities (retinal image)
certain primitives or place tokens are derived

• zero-crossings
• edges
• bars
• blobs
• terminations

• edge segments
• virtual lines
• groups
• curvilinear organisation
• boundaries

Consider a TV commercial using a reverse zoom

• initially a group of individuals is seen - they
  are assigned
• a visual token each
• as the camera rises, zooms backwards, it can be
  seen that
• the people form various groupings (a token for
  each group)
• at its greatest height the camera reveals the
  people as a letter
• - seeing each letter as a coherent whole
  implies that it must
• be represented in the visual system (as a new
  token)

M A R R
43

• primal sketch

M A R R
MARR
During the development of the primal sketch
groups of adjacent tokens having a common
property, such as orientation, are replaced by
"level one" tokens representing this common
property. Then, if there are whole groups of
similarly oriented level one tokens, these are
used to construct boundaries between parts of the
full primal sketch.
The question arises as to how the visual
primitives of the primal sketch are actually
extracted. Let us examine the primitives known
as zero-crossing.
M A R R
44

• primal sketch

M A R R
MARR
zero-crossings
Important information about a shape and its
orientation comes from edges, contours and
boundaries that is from areas in the image where
intensity values are changing rapidly
z
M A R R
intensity change
first derivative
second derivative
45

• primal sketch

M A R R
MARR
zero-crossings
essentially act as spatial frequency filters
which may be tuned to different scales to capture
information on edges
For example hold up the handout in front of you
and face the side walls. Close your eyes tight
and slowly open them a very small amount to view
the page. The page will be seen as a bright area
against the wall and within this brightness will
be a grey area. This grey area on closer
inspection will be seen to be a series of grey
blocks (paragraphs) and on widening the eyes more
the blocks are seen to a series of black and
white horizontal stripes and finally the black
stripes can be resolved into individual words or
letters depending on our level of focus and
concentration.
It appears that if several different spatial
filters agree on the position of a contour in an
image, then an edge in the real world exists.
M A R R
A series of filters (zero-crossings) have been
processed
46

• primal sketch

M A R R
MARR
zero-crossings
the mechanism
there are cells in the retina and the lateral
geniculate nucleus which exhibit receptive field
properties - activity of the cells can be shown
to reflect patterns of stimulation of groups of
retinal cells
receptive fields are organised in various ways
and shapes
M A R R
zero-crossing primitives correspond to On-centre
and Off-surround receptive fields
47
(No Transcript)
48
micro-electrodes inserted into the visual pathway
can pick up impulses from a single neuron
in this way one can discover which region of
retinal receptors is connected to that neuron
because patterns projected on one part of the
screen will make the cell respond with a stream
of impulses
the area of the screen that causes the neuron to
respond is called its receptive field
49
for about 50 of ganglion cells, light falling in
the very centre of the receptive field gives
on-responses while a spot of light limited to a
surrounding area actually suppresses activity of
the cell while it is turned on and causes an
off-response when the light is extinguished.
if parts of a neuron's receptive field are
illuminated, that cell gives a burst of impulses
either when the light turns on or when it turns
off.
50
small signal
large signal
Illumination of both the central and surrounding
regions of the receptive field by a large spot of
light causes a much weaker response than light on
the centre alone
appears white because an on receptive field is
stimulated by the white bar but the dark squares
are not stimulating off fields. At the
intersection, the white cross stimulates on and
off receptive fields, giving the grey shade.
51
The Hermann grid
conclusion from the Hermann Grid

• First, the retina is telling the brain mainly
  about the beginning and end of each retinal
  illumination
• Secondly, localised illumination is much better
  than diffuse light. Ganglion cells detect changes
  in the level of light - differences of
  illumination in time or space

52
If it is difficult to believe that nothing
stationery is visible, it is even harder to admit
that we simply cannot see uniform areas of light
and dark except by virtue of their edges, and yet
this is certainly true. It is even possible to
fool the brain into thinking that two identical
areas differ in brightness simply by creating an
apparent edge between them
Only the transition edge can be detected by our
ganglion cells, but we perceive the pattern as if
the intensity on each side continued uniformly
away from the edge.