PSY280 - 7. Depth Flashcards
Positivism - Plato
The world could be an elaborate hallucination
all we have to go on is what’s coming into our senses
matrix: world doesn’t really exist
2 assumptions
there is a real world to sense (realism)
geometry of the real world is Euclidean
Euclidean Geometry
parallel lines stay parallel
internal angles of a triangle always 180 degrees
objects don’t change size/shape as they move around in world
dictates physicality of our world
Euclidean Geometry
3D world is being projected into retina which is 2D and curved
retina is non euclidian
we need to reconstruct the world as euclidian
how do we take these noneuclidean inputs + reconstruct it
Binocular Summation
combo of signals from each eye in ways that make
performance on many tasks better than with either eye alone
2 non euclidian inputs - each diff
lose 1 you can still see
laterally situated: almost see 360 degrees - see huge proportion of world
Binocular Summation
visual field for humans - 190 degrees, 100 degree overlap - better chance for predators to find small fast moving objects
2 eyes > 1 eye for threshold for very dim light - lower - increased sensitivity
better visual acuity - vernier acuity
visual search: find it faster with 2 eyes
Stereopsis
binocular disparity: diff betw 2 retinal images use as cue depth
depth perception: 2 eyes with overlap, but 2 slightly diff image
fall on slightly diff locations on the 2 retinas
Stereopsis
perception of depth that we use by taking advantage of disparity - stereopsis - stereo vision
stereogram - take advantage of stereopsis
stereoscope: force 1 image to each eye - fool eye into thinking you are seeing the same image
disparity allows emergence of depth
Pictorial Depth Cues
Cues to distance present when 3D world is projected onto a 2D surface
standing at same orientation + distance as photographer: only point where there’s no distortion betw 3D image + picture - no diff in retinal inputs
Pictorial Depth Cues
everything distorted any other viewpoint
perception doesn’t feel distorted
taking into account orientation of viewing - orientation of picture
Orientation
can be taken into account when viewing a picture
taking into account angle, without context, image is distorted
perceptual system can compensate for distortion using contexts
anamorphosis
Using rules of perspective to create 2D image so distorted it looks correct only when viewed from special angle - the accidental viewpoint, but the viewpoint that is desirable to see intended image
anamorphosis
artists are reversing it
double portrait - Hans
anamorphic skull: hanging from hallway, come at it from the right, see the skull
occlusion
Objects in front obstruct view of parts of another object
more likely images are result of occlusion
present in almost every natural scene
occlusion
1 of most reliable depth cues
non-metrical depth cue - gives info about depth order, but not depth magnitude. can’t know if green triangle is tree in close distance/pyramid in far distance
occlusion
metrical cue - exactly how far object is from our viewing point
this is the only non metrical cue
relative height
Elevation comes into play when you can see horizon:
far from horizon: closer
close to horizon: far away
true for objects both above + below horizon
relative height
no horizon relative height is relative to observer’s visual field (rather than the horizon):
higher in visual field: far away
lower in visual field: closer
diff distances in ground plane - objects at diff heights in retina
relative size
When 2 objects are of equal size, the one that is farther away will take up less of the visual field.
based on experience, they are same size, we assume 1 that takes up most space in the retina is closer
familiar size
When you use prior knowledge of object size to estimate distance
atmospheric perspective
More distant objects appear less sharp + often have slight blue tint
implicitly know light is scattered in atmosphere
everything looks slightly hazy at farther distances
atmospheric perspective
scattering of light inversely proportional to wavelengths
slightly bluish tint - S wavelengths more scatter
light from sun sends white, red wavelengths come directly at you, blue light gets scattered more deflected everywhere making sky look blue
texture gradient
Elements equally spaced appear to be more closely
packed as distance increases
relative size: assume same approximate size so smaller is farther away
relative height: ground plane - higher up further aways
perspective convergence
convergence point is vanishing point
assumptions that outer edges are parallel, starting to converge toward back
Lines that are parallel in the 3-dimensional world appear to converge in a 2-dimensional image
motion parallax
When an observer moves, objects nearer observer move faster than more distant objects:
fence pickets: fast = close
farm house: slow = far
motion parallax
image of house moves a shorter distance on retina, so it looks like it’s moving more slowly as observer moves.
retina at position 1 + position 2
plot how far object has moved on retina
deletion & accretion
relative positions change as dog is moving
as moving - some things are being covered - deletion - others revealed - accretion
convergence
brain receives kinaesthetic info about eye position
How far eyes roll in (close)/out (far) gives info
far away, eyes roll out to be parallel
eyeballs move inward for something close
in both cases - keep same object on same foveal point
convergence
limit as to how much info it can give you - arm length or less: most parallel that they’re gonna be
accommodation
lens changes shape to bring light into focus on retina:
•thinner g far
•thicker g close
brain gets info from eye muscles that control accomodation
accommodation
knows when lens is thinner + thicker
accommodation + convergence coupled
useful within reaching distance, but don’t need 2 eyes for these
Binocular disparity
examining corresponding points on the retinas.
overlap retinas - identify corresponding points
By fixating the green gummie bear, the image falls on corresponding F points on the retinas
F = fixation + fovea
Binocular disparity
every time you fixate - imaginary arc that passes through fixation: anything on horopter falls on corresponding points on the retina
Objects not on horopter fall on non-corresponding points on retinas
angle of disparity
further from horopter, greater angle of disparity
anything on the horopter = same distance from where you are fixated
angle of disparity
on non corresponding points means they’re not the same distance as you
gives you info about how far from horopter, but not whether its front or behind
angle of disparity
Objects in front of horopter in crossed disparity
In the right eye unfixated object is to the left of fixated object, vice versa
when fixating red, blue is in front of horopter
angle of disparity
Objects behind of horopter are in uncrossed disparity
In the right eye unfixated object is to the right of fixated object, vice versa.
how far based on angle of disparity
1830s
Retinal disparity sufficient to create the perception of depth
3D glasses - filters means you deliver diff images to each eye giving binocular disparity
Wheatstone: invented stereoscope
depth perception given by retinal disparity
taken advantage of knowledge with 3D images, viewmaster
how is the correspondence problem solved?
Free fusion requires you to decouple accommodation and convergence.
when converged on close objects lens becomes thicker
need to converge eyes + allow lens to become thinner
in either case, have to figure out which bit of image in left eye corresponds to which bit is presented in right eye
by matching up object identity
Bela Julesz
developed random dot stereograms, which have retinal disparity without pictorial cues.
but produced random dot stereograms which have retinal disparity
matching 2 disparate images is how we perceive depth
even with no features, disparity is enough of a cue
Magic Eye (autostereograms)
using retinal disparity, but trick is to look through image
roll eyes out - produce slightly diff images - disparate images
Binocular depth neurons
receive inputs from both eyes; receptive field on one retina is slightly adjacent to corresponding point on other retina
has to be in V1 - that’s where you start getting input from both eyes
Binocular depth neurons
in a, b, c fixation point is the same
when person is fixated at dot, object slightly in front of horopter produces angle of disparity
red neuron:
responds best to stimulus closer to and slightly to the right of fixation: fall on receptive field of red neuron - maximum stimulation in front a little to right
blue neuron
responds best to stimulus further from + slightly to the left of fixation
behind of horopter + a little to the left maximum response from blue neuron
disparity tuning curve
animal models
stereograms allow you to deliberately control level of disparity in images you are presenting
using diff amounts of disparity - find tuning curve
each binocular disparity neuron has diff preferential disparity level
find optimal disparity
Stereo Sue
3-5% of ppl are stereo blind - don’t have binocular disparity neurons
can’t use binocular depth cues
strobismus - V1 don’t learn to integrate info from 2 eyes
thought to have sensitive period
she learned stereo vision in her 40s
never too late to learn stereo vision
perceiving distance is a big part of perceiving size
depth can give us a clue about size
if take up same amount of space in retina - one farther away is bigger
pictorial cues tell us that suv at back is farther away
Visual angle
angle subtended by an object on the retina
tan (Θ) = size distance
VA - irrespective of visual size: takes into account size + distance of viewing
Visual angle
really large object can take up small space if really far
small object can take same amount of space but closer
after images diff size on screen + paper
Emmert’s Law
farther away an afterimage appears, the larger it will seem
size-distance scaling: S = k(R x D)
s - perceived physical size dependent on k as a constant
Emmert’s Law
how much space on retina (R) x distance (D)
further so it seems large
on paper - perceived size is smaller, same space on retina
judgments about depth
if objects same physical size, but take up different amounts of space on the retina, MUST be at different depths (R) °VA = different (S) size = same (D) depth = ? S=RxD SL = 10°x10=100 SR = 5°x20=100
judgments about depth
hand isn’t changing size, but amount of space taking up on retina diff
taking up less space on retina, same space, but now larger distance
judgments about depth
if objects are at diff distances but take up same amount of space on retina, must be diff physical sizes (R) VA = same D depth = diff S size = ? S -R x D SR = 10 x 10 = 100 SB = 10 x 20 = 200 if at diff depths, then they can’t be the same size
size constancy
perception of an object’s size stays relatively constant, even when we view the object from diff distances
retinal space is smaller, but distance is increasing, so same size
Holway & Boring (1941)
Subjects matched physical size of the test circle sat in a hallway
Holway & Boring (1941)
each test circle 1 degree of VA, but varied physical size of object
closer, object had to be small, far, big object
they didn’t match VA, but physical size
also considered distance
Holway & Boring (1941)
unless depth cues were removed, subjects matched visual angle
1. all depth cues available
2. remove binocular depth cues
3. look at peep hole - remove motion depth cues
4. one eye, peep hole, curtains in hallway to remove pictorial depth cues
don’t need binocular to be good at perceiving depth
Scaling
can also be achieved based on familiar objects for comparison
Muller-Lyer Illusion
centre line is exactly the same length
misapplied size constancy scaling
inside corners are receding, outside corners are jutting out
S=RxD
seeing outward facing fin as inside corner
inside facing fin as outside corner
misapplied size constancy scaling
inward facing fin needs to be about 7% larger for noneuropean, need to be 20% larger for europeans
ppl in european civilizations deal with a lot of buildings
nomadic pops not really have experience, less likely to engage in this effect
can’t explain all of it because noneuropeans aren’t at 0
conflicting cues theory
2 (conflicting) cues to length:
•line length
•overall length of the figure
also works with similar, but slightly diff versions
Ponzo Illusion
manipulation of pictorial cues makes it seem as if objects are at different distances
S = R x D
Ames Room
manipulation of pictorial cues makes it seem as if 2 people are at the same distance.
Moon Illusion
moon looks larger when it is near the horizon than when it is overhead.
°VA is constant in the same night, and doesn’t
vary much over the course of a year
apparent distance theory
assumptions about the shape of sky makes it seem as if the moon is at different distances
flattened bowl effect
assume sky overhead closer to us than sky on horizon
same R diff D, larger D when on the horizon
angular size contrast theory
sky is huge But the moon is only a small fraction of that… it must be small
appears larger when close to large object
sky makes moon look smaller