4. Perceiving and recognizing objects
What and where pathways
Cells in V1 basic features of visual images (edges/lines of specific orienta°, mo°, size etc.) Small &
precise receptive fields. Beyond V1 extrastriate cortex (set of visual areas lying outside primary visual
cortex. Basic, local properties = pulled out of img by early stages of visual cortex, but sophisticated tasks
like object recognition = subsequent processing +++ & number of visual processing areas required +++.
Extrastriate cortex = region of cortex bordering V1 & containing multiple areas involved in visual
processing.
V2 rf = properties important for object perception. V2 cells “boundary ownership”. Might respondboundary ownership”. Might respond
to dark edge of an object but not to a dark edge created by a light object on a dark background.
From extrastriate regions of occipital lobe two pathways for visual info.
Where pathways parietal lobe, dorsal pathway, location of objects in space & actions required t
interact w/ them (moving hands, eyes), role in attention.
What pathway temporal lobe, ventral pathway, explicit acts of OR. Evidence: lesion studies.
Move down TL receptive fields go bigger. Distinction btwn two = sometimes blurred.
Agnosia = failure to recognize objects in spite of the ability to see them.
Inferotemporal cortex = in the lower portion of the TL, important in OR.
Hierarchical model of visual perception (small rf s & simple features of visual cortex = combined w/ more
and more complexity as moving from the V1 to the IT). IT cortex: connec°s w/ parts involved in memory
forma° (hippocampus) IT cells need to learn their rf properties. Rf properties of V1 genetic code?
but not the case for IT cells. These cells have the plasticity required for this. Very specific objects are
recognized by these cells (may fire to the Tour de Pise but not to the Eiffel Tower).
Visual system has different pbs to solve (face processing…) & have specialized modules for working on
different pbs.
Some processing regarding categoriza° of objects super-fast. Sometimes, not much feedback from
higher visual or memory processes possible to do some rough OR on the basis of the first wave of
activity as it moves form retina to V1 to extrastriate cortex & beyond.
Feed-forward process = carries out a computa° (ex: OR) one neural step after another, w/out need for
feedback from a later stage to an earlier stage.
Sum: 2 pathways emerge from visual cortex. What moves through a succession of stages; building a
representation of an object out of the very specific, localized spots, lines and bars that interest the cells
in the retina, LGN & V1.
The problems of perceiving and recognizing objects
Recognize a house first, gather basic visual features. None of the simple cells of the V1 see a house.
They collect local features like horizontal, vertical and oblique lines. These local features need to be
assembled into a house. Objects that are against the walls of the house: why don’t we consider them
part of the house? We have features that combine feature into objects = one of the tasks that defines
middle vision (opposed to low-level vision, which is the subject of chapters 2 & 3).
Middle vision = stage of visual processing that comes after basic features have been extracted from the
img (low-level, or early vision) & before object recognition & scene understanding (high-level vision).
Middle vision
Goal of mv = organize elements of a visual scene into groups that we can recognize as objects.
, Finding edges
How do we decide what belongs to which object? Objects abut & overlap other objects simple
connectedness wouldn’t work. When part of the object has no edge (same luminance than background
at some places), it doesn’t bother our visual system (even hard to see the gap). The vs knows that gaps =
accidents of the lightnings & it fills in the contour. Early stages vs figures out which edges mark
boundaries of objects & which represent surface features.
Illusory contours perceived because they are the best guess about what is happening in the world at
that location. Illusory contour = contour perceived even though nothing changes from one side of it to
the other in an image.
Rules of evidence
Structuralism = school of thought believing that complex objects or perceptions could be understood by
analysis of the components. Structuralists argued that perceptions = sum of atoms of sensation (bits of
colour, orientation etc.). An illusory contour challenges this view (an extended edge is seen bridging a
gap where no local atom of ‘edgeness’ can be found).
Flaws in structuralism Gestalt school. Gestalt = ‘form’. School of thought perceptual whole =
greater than sum of parts.
Gestalt grouping rules describes the vs’ interpretation of the raw retinal image. Gestalt grouping
rules = set of rules describing which elements in an image will appear to group together.
Visual system = trying to make sense of the inputs from the early stage of visual processing. These rules
reflect regularities in the world.
We tend to see similarly oriented lines as part of the same contour.
If two contour elements are close to colinear likely to come from the same contour (good
continuation principle)
Perceptual ‘committees’
Closure = Gestalt principle a closed contour is preferred to an open contour.
Occlusion = why does an edge stop in an image? Because something else gets in the way & blocks our
view vs sees an illusory contour.
Texture segmentation and grouping
Texture segmentation= carving an image into regions of common texture properties.
Similarity = Gestalt grouping rule tendency of two features to group together will increase as the
similarity btwn them increases.
Proximity = “boundary ownership”. Might respond-“boundary ownership”. Might respond as the distance btwn them decreases.
Parallelism = rule for figure-ground assignment parallel contours = likely to belong to the same figure.
Symmetry = rule for figure-ground assignment symmetrical regions = likely to be seen as figure.
Camouflage
Principles sued to help us find objects in the world can be used to hide them. Camouflage = getting
your features to group w/ the features of the envt persuade an observer that your features do not
form a perceptual group of their own.
Perceptual committees revisited
Middle vision collection of specialists, each w/ a specific area of expertise & opinions about what the
input means. Goal emergence of a single answer out of this diversity of opinions. This “boundary ownership”. Might respondcommittee” =
interconnected set of neurons that takes input produces an output = can be read ad the committee
decision by other processes elsewhere.
What and where pathways
Cells in V1 basic features of visual images (edges/lines of specific orienta°, mo°, size etc.) Small &
precise receptive fields. Beyond V1 extrastriate cortex (set of visual areas lying outside primary visual
cortex. Basic, local properties = pulled out of img by early stages of visual cortex, but sophisticated tasks
like object recognition = subsequent processing +++ & number of visual processing areas required +++.
Extrastriate cortex = region of cortex bordering V1 & containing multiple areas involved in visual
processing.
V2 rf = properties important for object perception. V2 cells “boundary ownership”. Might respondboundary ownership”. Might respond
to dark edge of an object but not to a dark edge created by a light object on a dark background.
From extrastriate regions of occipital lobe two pathways for visual info.
Where pathways parietal lobe, dorsal pathway, location of objects in space & actions required t
interact w/ them (moving hands, eyes), role in attention.
What pathway temporal lobe, ventral pathway, explicit acts of OR. Evidence: lesion studies.
Move down TL receptive fields go bigger. Distinction btwn two = sometimes blurred.
Agnosia = failure to recognize objects in spite of the ability to see them.
Inferotemporal cortex = in the lower portion of the TL, important in OR.
Hierarchical model of visual perception (small rf s & simple features of visual cortex = combined w/ more
and more complexity as moving from the V1 to the IT). IT cortex: connec°s w/ parts involved in memory
forma° (hippocampus) IT cells need to learn their rf properties. Rf properties of V1 genetic code?
but not the case for IT cells. These cells have the plasticity required for this. Very specific objects are
recognized by these cells (may fire to the Tour de Pise but not to the Eiffel Tower).
Visual system has different pbs to solve (face processing…) & have specialized modules for working on
different pbs.
Some processing regarding categoriza° of objects super-fast. Sometimes, not much feedback from
higher visual or memory processes possible to do some rough OR on the basis of the first wave of
activity as it moves form retina to V1 to extrastriate cortex & beyond.
Feed-forward process = carries out a computa° (ex: OR) one neural step after another, w/out need for
feedback from a later stage to an earlier stage.
Sum: 2 pathways emerge from visual cortex. What moves through a succession of stages; building a
representation of an object out of the very specific, localized spots, lines and bars that interest the cells
in the retina, LGN & V1.
The problems of perceiving and recognizing objects
Recognize a house first, gather basic visual features. None of the simple cells of the V1 see a house.
They collect local features like horizontal, vertical and oblique lines. These local features need to be
assembled into a house. Objects that are against the walls of the house: why don’t we consider them
part of the house? We have features that combine feature into objects = one of the tasks that defines
middle vision (opposed to low-level vision, which is the subject of chapters 2 & 3).
Middle vision = stage of visual processing that comes after basic features have been extracted from the
img (low-level, or early vision) & before object recognition & scene understanding (high-level vision).
Middle vision
Goal of mv = organize elements of a visual scene into groups that we can recognize as objects.
, Finding edges
How do we decide what belongs to which object? Objects abut & overlap other objects simple
connectedness wouldn’t work. When part of the object has no edge (same luminance than background
at some places), it doesn’t bother our visual system (even hard to see the gap). The vs knows that gaps =
accidents of the lightnings & it fills in the contour. Early stages vs figures out which edges mark
boundaries of objects & which represent surface features.
Illusory contours perceived because they are the best guess about what is happening in the world at
that location. Illusory contour = contour perceived even though nothing changes from one side of it to
the other in an image.
Rules of evidence
Structuralism = school of thought believing that complex objects or perceptions could be understood by
analysis of the components. Structuralists argued that perceptions = sum of atoms of sensation (bits of
colour, orientation etc.). An illusory contour challenges this view (an extended edge is seen bridging a
gap where no local atom of ‘edgeness’ can be found).
Flaws in structuralism Gestalt school. Gestalt = ‘form’. School of thought perceptual whole =
greater than sum of parts.
Gestalt grouping rules describes the vs’ interpretation of the raw retinal image. Gestalt grouping
rules = set of rules describing which elements in an image will appear to group together.
Visual system = trying to make sense of the inputs from the early stage of visual processing. These rules
reflect regularities in the world.
We tend to see similarly oriented lines as part of the same contour.
If two contour elements are close to colinear likely to come from the same contour (good
continuation principle)
Perceptual ‘committees’
Closure = Gestalt principle a closed contour is preferred to an open contour.
Occlusion = why does an edge stop in an image? Because something else gets in the way & blocks our
view vs sees an illusory contour.
Texture segmentation and grouping
Texture segmentation= carving an image into regions of common texture properties.
Similarity = Gestalt grouping rule tendency of two features to group together will increase as the
similarity btwn them increases.
Proximity = “boundary ownership”. Might respond-“boundary ownership”. Might respond as the distance btwn them decreases.
Parallelism = rule for figure-ground assignment parallel contours = likely to belong to the same figure.
Symmetry = rule for figure-ground assignment symmetrical regions = likely to be seen as figure.
Camouflage
Principles sued to help us find objects in the world can be used to hide them. Camouflage = getting
your features to group w/ the features of the envt persuade an observer that your features do not
form a perceptual group of their own.
Perceptual committees revisited
Middle vision collection of specialists, each w/ a specific area of expertise & opinions about what the
input means. Goal emergence of a single answer out of this diversity of opinions. This “boundary ownership”. Might respondcommittee” =
interconnected set of neurons that takes input produces an output = can be read ad the committee
decision by other processes elsewhere.
