Lecture 2:

Lecture 5:
Object perception

PSYC 2220 M: Sensation & Perception I

CB 121, Fridays, 11:30 AM
Instructor: Raphael Gastrock
Midterm 1 next week
• Friday (Feb. 14th) at 11:30 AM – CB 121
• 2-hours
• Bring writing materials (pen/ pencil)
• Student ID (passport or YORK ID)
• Need your student number
• Mobile York ID is fine
• Q&A session after today’s lecture
Object perception
• When looking out into the world, what do
you name the things you see?
• We do not usually see center-surround
patterns, or sine wave gratings
• We see spaces, objects, surfaces, colors,
lights, motion
• We are good at perceiving even unfamiliar
objects
Object perception
• How it works in the brain is not well and
fully understood yet
• There are some ideas and theories
Physiology
• Recall that V1 has receptive fields that respond
best to edges/ lines of specific orientation,
motion, and size
• Extrastriate cortex = region of cortex
surrounding V1, where its different areas are
involved with different visual processing
Macaque visual areas
Human visual areas
Human visual areas
Physiology
• V2 area starts to show preference for properties
important for object perception
• Border ownership
• The edge of an object belongs to that object
• Notice how V2 receptive fields would then differ
greatly from receptive fields in the retina, LGN, or
V1
What and Where pathways
• What pathway
• Ventral pathway to
temporal lobe
• Object recognition
• Where pathway
• Dorsal pathway to
parietal lobe
• Location, actions
Inferotemporal (IT) cortex
• From simpler cells, object
recognition is built
hierarchically as you move
into IT
• Receptive fields can span
half the visual field
• Cells respond to particular
types of objects: hands,
faces, sticks
• “Grandmother cell”
• Homologous regions in
humans
• “Jennifer Aniston cell”
Inferotemporal (IT) cortex
• Do “grandmother cells” likely exist in the brain?
• If grandmother cells die, will you no longer recognize your
grandmother?
• Do we have a cell for every possible object out in the
world?
• We have different grandmothers, so do we have different
grandmother cells?
• “Grandmother cells” are more likely to be object-selective
cells that respond to different things, and play a role in a
much larger network of cells
• Highlights how the human brain can be specialized for
different stimuli
Other specialized brain
areas
• Fusiform face area (FFA)
• Responds to human faces
• Extrastriate body area (EBA)
• Responds to images of the body, other than the
face
• Parahippocampal place area (PPA)
• Responds more to images of places
• Visual word form area (VWFA)
• Responds to images of written words
• Mostly in left hemisphere, near FFA
Evidence for specialized
brain areas
• fMRI studies of patients with stroke
• Lesions = region of brain that is injured/ damaged
• Leads to agnosia
• Ability to see object, but cannot recognize the
object
• Prosopagnosia = inability to recognize faces
Reverse-hierarchy theory
• Feed-forward processing (flow of
information from retina > LGN >
striate cortex > extrastriate
cortex)
• Actually occurs very fast
• EEG data show that we can
distinguish animals from non-
animals within 150 ms
• This is not enough time for
feedback from later visual
areas to earlier ones
• Allows for coarse object
perception
• Feedback processing
• From later or higher visual
areas allows for finer
Perceiving and recognizing
objects
• We perceive
all of these as
an “elephant”
• But we also
perceive each
elephant as
being made
up of different
materials
• low-level vision: spots of light, edges, sine waves,
etc.
• Retina, LGN, V1
• high-level vision: specific learned objects and
patterns
• E.g. identifying an “elephant” when you see one
• IT cortex
• mid-level vision:
• Vague area between low and high-level vision
• Not clearly defined, but is obviously useful
• E.g. border ownership (V2)
• contours, groups, surfaces, figures, borders, etc.
Mid-level vision: Contours
• Perception of contours is a
good first step in object
recognition
• Shape of an object,
defined by its edges
• But this presents some
problems:
• We can see that
objects’ edges can
overlap with each other
• Which edges belong
together?
• What about
disconnected or
missing edges?
• Kanizsa figure
• illusory contours
• Perception of a
contour, even if one
side of the image
has no change from
another side
• Visual system’s best
guess about what is
happening in the
world
• Illusory contour as the
visual system’s best
guess
• If the line stops,
something must be
occluding it
Schools of thought:
Perception
• Structuralism
• Percepts are built up from “perceptual atoms”
• Combining of edges, patches of color,
orientation, etc.
• Problem: illusory contours challenge this view
• How can we perceive a continuous edge, even
if there are no “edges” to be found?
• Gestalt psychology
• Percepts are more than the sum of their parts
• Gestalt grouping rules
• Reflect that there are patterns or certain
configurations in the world
Gestalt grouping rules
• Good continuation
• Two objects will tend to group together if they
seem to lie on the same contour
Gestalt grouping rules
• Good continuation
• But, context matters
• While Gestalt rules can shape how we see the
world, the world can also shape what we see
Gestalt grouping rules
• Good continuation
Gestalt grouping rules
• Similarity
• Objects similar to each other are more likely to
be grouped together
Gestalt grouping rules
• Similarity
• But, this falls apart with a combination of
different features
Gestalt grouping rules
• Proximity
• Items near each other are more likely to be
grouped together
Gestalt grouping rules
• Common region
• Objects within a closed region are grouped
together
Gestalt grouping rules
• Connectedness
• Objects that are visually connected are
grouped together
Applications of grouping
rules: Camouflage
• The principles we use to help us find objects can
also be used to hide them
• Get features of object to group with features of
the environment
Camouflage
• The principles we use to help us find objects can
also be used to hide them
• Get features of object to group with features of
the environment
• Can also be features that confuse viewer, rather
than hiding the object
Applications of grouping
rules: Texture
segmentation
• Carving an image into regions of common texture
properties
Perceptual committees
• Percepts depend on multiple pieces of information
• Each “committee” argues about how an object is
to be perceived
• Here, the committees argue for either grouping
by proximity or similarity
• But which is correct?
Perceptual committees
• Perceptual committees come up with a single
percept, even if presented with ambiguous
images
Perceptual committees
• Perceptual committees take into account
accidental viewpoints
• Viewing position that produces some regularity
in the visual image, even if it is not present in
the world
Perceptual committees
• Perceptual committees take into account
accidental viewpoints
• Viewing position that produces some regularity
in the visual image, even if it is not present in
the world
Nonaccidental features
• Feature of an object that is not dependent on the
exact viewing position of the observer
• Helps perceptual committees deal with occlusion
Figure & ground
• Distinguishing objects in the foreground (figure)
from objects in the background (ground)
• Figure-ground assignment is the process of
determining which parts of an image belong to
figure or ground
Figure & ground
• Principles to determine figure from ground
• Surroundedness
• If one region is entirely surrounded by the
other, than the surrounded region is likely the
figure
• Size
• Smaller region is likely to be the figure
• Symmetry
• A symmetrical region is likely to be seen as the
figure
• Parallelism
• Regions with parallel contours are likely to be
seen as the figure
• Relative motion
• How surface details move relative to an edge
Bayesian models
• Obviously, we don’t really have perceptual
committees
• We can mathematically model the probability (P)
that the world is in a given state (A), given a
particular observation (O)
• Bayes’ theorem: P(A|O) = P(A) x P(O|A)/P(O)
• Prior probability = probability of an event
before you collect observations
• Uses your knowledge about the conditions of
the world
• E.g. We are more likely to have snowy
conditions in January than July
• Priors are combined with observations
• How consistent your observation is with your
prior
Bayesian models

E + 3 is unlikely – needs
to line up (accidental
viewpoint)
Could be equally likely
but inconsistent with
observation

Equally likely and

consistent observation
Heider & Simmel (1944)
• Example of how our brain (or perceptual
committees) can assign different meanings to
what we perceive
• Perception depends on context
Best of luck with the midterm!