Lec 10-

Question 1

Gestalt psychologists said what
Full primary sketch uses waht

Answer 1

The whole is greater than the sum of the parts. Though individual lines may be different they all group together to produce long horizontal bar. Not just features in isolation.

Full primal sketch Marr used gestalt laws of perceptional organisation and larger structures boundaries and regions made explicit using gestalt like grouping rules of the primitives in the raw primal sketch.

Question 2

Full primal sketch some info
What does it provide a list of, goal of full primal sketch and how is the task of recognition made esaier

Grouping few lines into blob cell reveals what. Where are these properties present.

New symbols and higher level symbol. Old lower level symbols.

Answer 2

The Raw Primal Sketch provides a list of symbols like edges and line terminations. The goal of the Full Primal Sketch is to group these primitive elements into more meaningful clusters: the resulting compound descriptive elements have new properties, which may make the task of recognition much easier.

For example, ‘spots’ have properties such as position and size. A line of spots can be described in terms of their individual positions but, if the spots are grouped explicitly into a new line symbol, the symbol has the new explicit properties of length, width and orientation

Similarly, grouping a few lines into a new blob symbol reveals new properties such as shape. In each case, the new properties are actually present in the low-level elements, but the process of forming the higher level symbol makes these properties explicit. The new symbols are thus much more useful for further processing. They do not completely replace the old lower-level symbols but just provide them with a new ‘handle’: the original descriptions of the low-level elements are still available if needed.

The processes of visual grouping were studied extensively by the Gestalt psychologists , and one interpretation of their famous dictum “the whole is greater than the sum of its parts” is in terms of the higher order descriptions that emerge when primitive descriptive elements are combined into higher-order symbols.
The Gestalt psychologists formulated a set of grouping rules to describe typical human perceptions.

(All individual lines diff but group together to produce long horizontal bar so grouping= larger structure,. Full primal sketch used gestalt laws of perceptional organisation.

Question 3

Grouping rules
Proximity

Answer 3

Proximity- groups into rows rather than columns as nearest ones are to right not underneath. Reverse is true vertical if vertical; closer.

Similarity- can override proximity here though vertical gaps bigger than horizontal gaps we should see it in rows and we would if they weren’t coloured by if they are then we see similar objects and the figure is organised into columns of similar elements eg brightness. But proximity can sometimes override similarity.

Closure- closure of elements now. Symmetry is important for this. Baso outlines of objects you see the bit in the space the gap of the outline so closure of elements top part with bottom part for one complete figure for hidden message. Not individual parts.

Closure can also make us see contours that arent actually there bc our brains are trying to make closure so illusory contours as brain fills gaps. Whole greater than the parts so we see panda for example eventhough white is same we see that illusory contour as brain fills space between his ears

Question 4

Good continuation
Symmetry

Answer 4

Continuing on so we see them together.

Symmetry- makes a greater whole but we do see both individually as well. So binds the two parts together but in a greater whole. When w break that symmetry now its like two unconnected objects so its important in binding things together. Clearly defined axis of symmetry= stronger orientation.

Question 5

A modal completion

Answer 5

Not exactly grouping rule. When parts of images we cant see are completed, we suppose this character has legs when cut off but we dont see it so binds things that are visual are there with what is implied to be there sp groups together in that way.
Works in harmony with good continuation as we see two sticks crossing behind tree trunk and we amoddally complete missing part of image behind the tree. May be wrong

Good continuation= simpler overall image

Question 6

Common fate

Answer 6

The tendency to group things moving in the same way
If staying still cant see it as well if moving together probably have a common object.

Segments objects from scattered distinguishing background so when moving helps group together so easier to see but stationary cant eg moth. So prey= stationary when camouflaged so not distinguished from background.

Question 7

The marroquin pattern

Answer 7

Spontaneous organisation and recognisation.. your brains trying to group things so it switches between things as it cant fixate on a set structure so lots of competing structures with different grouping rules. It cant settle on a solution so it cycles around all the solutions.

This shows these rules are always operating. Grouping by similarity allows us to do this

Question 8

Figure ground segregation and competition between grouping rules

Answer 8

An important stage that follows grouping is the assignment of figure and ground. By ‘figure’, we usually mean, object, though it could be a collection of objects (e.g. a pile of trash, where the objects are grouped together) or a flat 2D image (e.g. a painting), where the content is enclosed by the frame. By ‘ground’, we mean the space between the objects, sometimes called background.

In the main, the visual system is very good at this—look around you and consider the various challenges that face the visual system in performing this task. On occasions though, the decision can be ambiguous, as in Rubin’s vase in Fig 6. This is another example of an ambiguous figure—is it a vase or two faces? Note that figure and ground are being assigned differently in each case. This task of border ownership (whether the boundary is owned by a white region or a black region) is thought to be performed by cells in V2 (Zhou et al, 2000). So is the boundary owned border ownership by the white teh vase or the faces the black…. V2 cells

Question 9

Competition between grouping rules

Answer 9

Closure proximity work together now if we add them together and close them off then closure wins.

Proximity vs similarity rows and columns if pull them apart proximity wins but sometimes similary wins. Cant put them in hierarchy

Question 10

From the raw primal sketch to full primal sketch

Answer 10

Marr’s technique for getting from the Raw Primal Sketch to the Full Primal Sketch is really just an implementation of these simple grouping principles.

But whereas the Gestalt psychologists simply described grouping rules, Marr provides some (admittedly rather vague) theoretical justification by pointing out that image elements that share any physical similarity (e.g. scale, brightness, orientation) are likely to arise from some common cause in the external world (e.g. a group of points moving together are really very likely to belong to the same object). In other words, Marr linked the grouping rules, which apply to the proximal stimulus, back to the distal stimulus.

Question 11

Pragnanz

Answer 11

General principle tp all grouping rules
Law of good figure= simplicity.

Simplciity= compared with loghitness contrast and orientation.

Question 12

Texture

Answer 12

Simple images dont imply pictorial relief just lines on screen but it has visual texture. All similar elements group by similarity to form 2 different textures= texture boundary between them. Another way for figure grounds egregation.

Texture is a key aspect that emerges from grouping in visual perception. Grouping line elements by orientation can quickly reveal distinct regions, a process known as “preattentive” segmentation, indicating rapid visual processing. Texture descriptions aid in recognition and depth perception, and texture boundaries, highlighted in the Full Primal Sketch, help identify object boundaries not visible in the Raw Primal Sketch.

Experiments with texture patterns help explore human vision’s grouping processes. The ability or inability to segment textures may reflect natural properties. These experiments prompt us to consider the brain’s wiring and the role of filtering processes by simple and complex cells in texture segmentation.

Question 13

Criticisms

Answer 13

Visual search we see odd item pop out= pop out phenomenon and rest segment away so takes longer= conjunction search.

We can do experiments and manipulate no of distractors which is independent variable and measure reaction time but still it doesnt slow down, greater no of distractors we have to search more.

Grouping and parallel search can be driven by depth info so can get back down from 2 and 1/2d sketch to primal sketch either way strict feed forward arrangement with each stage solving different tasks cannot be true

Question 14

Criticisms once again

Answer 14

In Marr’s scheme, grouping takes place in the primal sketch based on 2D retinal coordinates, Fig 7, the odd-item out is defined according to its three dimensional geometry, implying a deeper level for grouping than is often supposed.

Note that the number and orientation of lines [i.e. the primitives in the raw primal sketch] is identical for all of the items in Fig 7, so the grouping that allows the odd-item out to be recognised has access to information of an order higher than a simple edge description.

Question 15

What may the grouping rules be compared to Mars scheme
When will pop out occur

In the visual search paradigm

Answer 15

Grouping rules may be deeper than supposed in Mars scheme bc stimuli can be created in which perception of depth plays an important contribution to grouping.

Pop out will occur so long as basic features for the target are arranged differently from the distractors.

Visual search- if time taken to search for a target does not depend on number of distractors in the display the graph of results will be a flat line as (doesnt increase or decrease visual search time based on searching for target as it doesnt depend on distractors so same reaction time)

Question 16

Is pragnanz associated w gestalt psychologists
What about 2 and 1/2D sketch is true

Answer 16

Pragnanz is not associated with gestalt psychologists.

2/12/d d slektch- the types of things made explicit at this representation level would be use for navigation as depth cues so surface properties can be valuable so judging distances and avoiding obstacles etc. 2 and 1/2 d sketch describes only the visible parts of the scene.

Question 17

Correct perception of the size of unfamiliar object requires

Answer 17

That the visual scene contains reliable depth cues.

Question 18

If image features dont group

Answer 18

They segment

Question 19

Does grouping involve top down and bottom up processing

Answer 19

Does grouping involve top down and bottom up processing- yes

Question 20

Mars 2-1/2 d sketch. Wat did we see in. What’s the problem

Answer 20

Marr’s 2-1/2D Sketch describes surface properties of the image (with no regard for the objects for which those surfaces are part of) and includes a depth map of the 3D layout of the world from the current viewpoint (i.e. it is viewer centred).

Even though images are flat there are many ways to recover depth information. Collectively, these are called depth cues (some of which are shown in Fig 1) and include such things as convergence and stereopsis from binocular vision. However, as shown in Fig 2, even with only the type of information potentially available from a single (i.e. monocular), black and white Full Primal Sketch, we can make good sense of depth using static pictorial depth cues

Question 21

Depth cues what is it

Answer 21

In vision, a cue is something that allows us to make inferences about the stimulus. Thus, in the case of visual depth cues, certain aspects of the retinal image allow inferences to be drawn about (relative) depth and the order of objects in the 3D world.

Computational theories of vision are really just a formal elaboration of the properties of the most useful cues.

Question 22

Consider a cube info in the 2-1/2 d sketch

Retinal image is flat but works is 3d

What 2 categories do depth cues fall into

2 types of visual cues

Answer 22

Occluding contours- segments object from background
Surface orientation discontinuities- made explicit at this stage,
Visible surfaces- only seen and behind it not explicit
Surface orientations so slant and tilt
Distance of each surface from observer. Estimate of distance.

3rd dimension is embedded in 2d retinal image.

Visual and oculomotor which splits into convergence and accom. Convergence angle can tell us how far away the object s as our eyes swing in together as objects come near us. Accomodation again same thing if we can get feedback from states of the lens.
Oculomotor- one surface or object at a time and only useful over short distances and minor source of depth info. So fairly minor.
Visual cues are more valuable.

Monocular and binocular. Monocular contains visual cues and cue doesn’t require comparison so closing one eye monocular cues to depths till available

Question 23

Static pictorial cues

Answer 23

Monocular. Dont involve motion. These determine how stationary 3d objects appear when projected onto flat 2d image. Major category of depth cue.

Question 24

Elevation

Answer 24

Pictorial cues include elevation (height) in the visual

For things above the horizon, more distant objects appear progressively lower in the image (this is not really apparent in Fig 2, but you could ‘sketch-in’ a couple of flying swallows to get the idea!).

Thus, vertical position in the image can provide an important cue to depth. Note, however, that just because an object is higher in the image and below the horizon, it does not mean that it has to be further away, just that it probably is. This is the essence of a cue—typically, it is something that generally points us in the right direction.

Question 25

linear perspective

Answer 25

One of the most obvious pictorial cues to depth is that of linear perspective: parallel lines in the 3D world that recede in depth project to converging lines in the 2D image (see Figs 4). Train tracks.

Converging image lines are thus a potential cue to depth. Notice, too, that they can provide important cues to 3D boundaries. Indeed, although the central line in Fig 4 is in fact straight, one can almost ‘feel’ it kink halfway up the figure.

Question 26

Texture

Answer 26

Texture is one of the main properties to emerge at the level of the Full Primal Sketch and, as pointed out by Gibson, texture gradients provide an important cue to depth (see Fig 5)—

as a textured surface recedes from us, the texture becomes more dense and the texture elements become smaller in the 2D projection. This suggests that in addition to simple things like texture density, more complex descriptions like gradients of density (i.e. how texture density changes over the retinal image) should also be made available in the Full Primal Sketch.

Question 27

size

Answer 27

Closely linked to texture gradients is size. In the real world, an object of fixed size subtends a smaller visual angle when it is further away from us.

Thus, retinal image size is itself a cue to depth—the smaller the retinal image of an object, the further away it is. In Fig 6, the disk on the left appears to be further away than the disk on the right.

Question 28

Aerial perspective or atmospheric perspective why does sky appear blue

So what do distant objects look like 1 and 2

Answer 28

Air full of particulate matter so more haze greater distance. As scatter occurs more.

Unlike the other pictorial cues, this one does not relate to image geometry . as an object moves further away from us there is a greater number of particles (e.g. dust & moisture) between us and the object (particularly in cities!).
This means that 1) light reflected off distant objects is scattered more than light reflected off nearby objects and 2) more light from the sky (known as skylight) is scattered into the line of sight for distant objects than for near objects.

Furthermore, the atmospheric scatter of skylight is greater for short-wavelength light than for long-wavelength light (this is why the sky looks blue).

All this results in distant objects appearing of lower contrast (because of 1), more blue (because of 2) and generally less colour saturated than near objects.

Laboratory studies have shown that a reduction in contrast and saturation really can make objects appear more distant, thus, atmospheric perspective is an effective depth cue.

Question 29

Occlusion p

Answer 29

When one object overlaps another, it is said to occlude it. This provides a cue to the order in which objects are set out in 3D space—the occluding object is nearest. Provides ordinal depth information. Ammodal completion and good continuation also important here giving us strong impression of depth. We know further away but dont know by how much so square on top of circle the one that is occluding or overlapping is closest to us

Kanisza triangle another example of occlusion white triangle on top appears to be closer to us.

Occlusion is a powerful depth cue, and there is evidence that it develops early. For example, infants as young as seven months can judge relative position using occlusion cues alone

-

Question 30

Shading
Robots vs humans
Highligh

Answer 30

Shading, shadows and specular reflections are good sources of local depth information, providing information about surface curvature. Flat image but shading can help us see where light source is so we see 3d relief.

But robots can also see shading cues and can use this cue better than humans so robot vision outperforms biological humans here

Highlights tell us quality of material. Highlight more glossy

Question 31

Binocular visual cues to depth
Motion parallax

Answer 31

One major source of depth information comes from the fact that the retinal images of the two eyes are slightly different.

So far, we have considered only static images, but when the observer moves relative to the world, the pattern of optic flow that passes over the retina provides information about relative depth, rate of approach and surface slant. This use of dynamic cues in depth perception will be dealt with more fully in the forthcoming lecture on motion.

Question 32

Familiarity

Answer 32

Object is familiar then under normal viewing conditions this can contribute to perception of size eg car so top down

Question 33

Size constancy revision
Paradox

Answer 33

The discussion delves into the complexities of how humans perceive object size and depth. It begins by acknowledging that as objects move farther away, their image size diminishes, yet our perception of their size remains relatively constant due to size constancy.

Traditionally, this constancy is believed to be achieved by estimating distance and adjusting our perception accordingly. However, there’s a paradox in that while image size can serve as a depth cue, perceived depth is also used to compute object size. The suggestion arises that texture and familiarity with object sizes may contribute to a general understanding of the depth plane, allowing us to gauge the size of less familiar or ambiguous objects by comparing the occlusion of texture elements.

This proposed mechanism aligns with Gibson’s theory of direct perception, wherein object size is inferred by the number of occluded texture elements, regardless of retinal image size or distance from the observer.

Question 34

Failures of size constancy

Answer 34

Occasionally, size constancy breaks down. One classical example is when objects are viewed from a high vantage point. Traditional arguments claim that the reason that cars and people look so tiny when viewed from a high-rise building is because the image is impoverished of its usual depth cues.

2) Another example is the sun and the moon which typically are perceived to be about the same size. Although the sun is considerably larger than the moon, it is also considerably further away. In fact, it turns out that the visual angle of the moon is the same as that of the sun. Thus, in the absence of appropriate depth cues perceived size is at least partially determined by visual angle.

3) A related example is the moon illusion. Experiments have shown that the moon appears larger when it is seen close to the horizon than when it is seen high in the sky. Of course, the visual angle of the moon is exactly the same in both situations, so why should it appear different? In fact, a whole textbook has been written on this phenomenon, but the traditional explanation is this. Ground based depth cues provide a strong impression of depth when the moon is viewed on the horizon but when viewed up above there are few depth cues and the moon is perceived as closer. Of course, if two objects have the same visual angle but one of them is perceived as further away, then the more distant object must be larger, just as is observed in the illusion.

4) Finally, size-constancy also fails when objects are viewed close-up. For example, close your left eye and splay out the fingers of your right hand. Now, with the middle finger of your right hand touching your right eye-brow gaze towards the palm of your right hand and notice the size of your fingers. Don’t they look huge!

Question 35

Is it true that due to atmospheric perspective distant objects look smaller

Answer 35

Is it true that due to atmospheric perspective distant objects look smaller- no they arent acc directly affecting size here.

Distant objects look closer than are when city dweller to country as less dust so the city dwellers atmosphere perception is still calibrated to city conditions so look closer.

Pxs must be calibrated for density of particulate matter in general area of country.

Question 36

Let’s talk about the depth cue elevation

Answer 36

For objects below the horizon if they are positioned below the horizon they are positioned higher in the vf so are interpret as being further away. Higher elevation

For objects above the horizontal positioned lower than vf so closer. Lower elevation.

Question 37

Facts about 2-1/2 d sketch

Answer 37

Only describes visible parts of scene- distance,s surface angles orientations, pose texture
Includes viewer centred depth map
Surface slant is made explicit
Stereopsis is important at this level
Adn things are made explicit here so useful for navigation.

Question 38

Depth cue, pictorial depth cue and occlusion

Answer 38

Depth cue= orientation
Pictorial= occlusion. Not motion parallax

For occlusion to work as a depth cue a modal completion must also be in operation.
It can also be monocular.
Occluding objects are seen as nearer than occluded objects.
It provides a cue only to the order of objects in the scene not their relative distance from the observer.
Involved in amodally completing the occluded part of an object cannot be based solely on info ab local edge junctions.
Kanisza triangle example of this occlusion using solid white triangle

Orientation cannot be used to gauge absolute depth but does provide indication of relative depth
It cannot indicate that one object is twice as far away as another.

Question 39

Depth is given by retinal image size bc distant objects always produce smaller retinal images than other objects closer to observer

Answer 39

Yes true generally smaller but retinal image size isnt sufficient alone to give depth information.

Question 40

do texture gradients indicate smooth changes in object size across viewed screen

Linear perspective-

Answer 40

Not always smooth changes.
It did functions s depth cue before reneiscance artists.

Question 41

Binocular vision and Stereopsis
Combining info from the two eyes together

Answer 41

BV any animal with two eyes and we get a binocular overlap from each of the eyes= prey mammals have very little bin as eyes placed on sides of heads for frogs so increases their fov.

Lack of bin overlap so little Stereopsis. Predators have eyes on front of head so they have lots of bin overlap so benefit from Stereopsis= fine 3d vision.

Fusion= Panums fusional space, finger demo at arms length= diplopia. Each eye has its own retinal image but we dont see double its combined together tossed single binocular perception. diplopia covered by suppression. Fusional space is larger for larger objects

Question 42

Why does the contrast of the world not change when you close one eye

Answer 42

Visual system doesnt sum images from two eyes rather it does binocular averaging. Binocular contrast gains control. When we can barely se anything the visual system doesnt average it sums.

Question 43

Anatomy and neurophysiology

Answer 43

Stereopsis, or depth perception, relies on comparing images from both eyes. Many prey animals have side-facing eyes for a wide field of vision, minimizing stereopsis, while predatory animals have forward-facing eyes, enhancing stereopsis for better depth judgment. To facilitate image comparison, the visual system has evolved so that images from the same part of the visual field are processed close together in the brain.

This is achieved by the optic chiasm, where each eye’s left retinal field projects to the left hemisphere and the right retinal field to the right hemisphere. Thus, both images of an object end up in the same brain hemisphere, aiding comparison. The LGN receives monocular input from each eye in separate layers, while the first binocular cells appear in V1, forming ocular dominance columns organized in alternating bands for each eye. These V1 binocular cells are the initial stage for stereopsis, with more binocular cells found in V2.

Question 44

Psychophysical evidence for bin mechanisms

Answer 44

Stereopsis aside, psychophysical evidence for a site receiving input from both eyes is abundant.

One simple example is the interocular transfer of the tilt aftereffect (TAE) (see Handout 5). If the adapting grating is presented to the left eye only (for say two minutes), and then the test grating is presented to the right eye only, the test grating is subject to an illusory tilt in a similar way to the conventional TAE.

The conclusion is that at some point in the visual pathway information must be combined from both eyes.

Question 45

Bin disparity and the horopter

Answer 45

In order to make comparisons between the two images, the visual system needs to establish which point on one retina corresponds with which point on the other.
reasonable starting point is to suppose that the correspondence is based on retinal geometry. For example, the corresponding point of the fovea in the left eye is simply the fovea of the right eye. We can generalise by saying that if a pair of retinal points (one for each eye) are displaced from the fovea by equal amounts in the same direction, then they are corresponding points.

Several horopter each defined by criterion. Eg empirical horopter curve looks same distance from observer as fixation point. When object is on the horopter it’s bin images fall on corresponding points on the two retinas.

Question 46

Disparity

Answer 46

Binocular disparity measures the distance between an image and its corresponding point on the retina. Distance between image and its corresponding point= measure of bin disparity.

Disparity is zero when images fall on corresponding points, meaning the object lies on the horopter. As an object moves away from the horopter, disparity increases, indicating the object’s positional deviation. The direction of this disparity shows whether the object is inside or outside the horopter. The magnitude of the disparity represents the size of the deviation from the horopter and is proportional to the physical deviation divided by the square of the viewing distance.

Thus, for the same magnitude of disparity, the physical deviation is greater when the fixation point is distant than when it is near. To accurately interpret depth, the visual system must scale disparity with the depth of fixation. This necessity is often overlooked in textbooks, and convergence cues may help in this scaling process.

Question 47

Visual direction

Answer 47

Previously, binocular disparity was discussed as the distance between image points and their corresponding points on the retina, measured in millimeters.

However, it’s more common to express disparity in degrees of visual angle. This is done by considering the visual direction of any point in space relative to the line of fixation for each eye.

The visual direction of a point is the angle between the zero direction (line of fixation) and the line from the center of the entrance pupil to the point. Disparity is the difference between these angles for each eye. If the eye were spherical, retinal distances and visual directions would be equivalent. Regardless of eye shape, the horopter defined by visual directions forms a perfect circle called the Vieth-Müller circle, which is accurate when points lie on a plane passing through the fixation point and entrance pupils. Other horopter definitions do not generally produce a circle.

Question 48

Fusion

Answer 48

Not everything is fused we can get two retinal images

The range of distances around the horopter over which the left and right retinal images of an object will fuse is known as Panum’s fusional space. once Panum’s fusional space is known, one can easily determine the range of disparities that will permit fusion, the outer reach of which is known as Panum’s limit. Recent research has shown that Panum’s limit depends upon the size of the stimulus elements (e.g. Smallman & MacLeod, 1994). For very small objects, such as fine wires, the limit is rather small (e.g. 0.1°, which is about ±9mm of object distance at a viewing distance of 57cm), but for larger objects (i.e. broadly spaced light and dark bars), the limit is considerably larger.

This goes some way towards explaining why, in general, we do not suffer as much diplopia as the classical Panum’s limit of 0.1° suggests. Another reason is that, even for small objects, Panum’s fusional limit is larger in the periphery than it is in the fovea, presumably reflecting the increase in receptive field size that is found with an increase in eccentricity

Question 49

Binocular rivalry

Answer 49

Diff images shown to each eye and sometimes image from one eye dominates and sometimes re and sometimes see mix of the two. Cycles around and around. When it blends it changes.

Eyes fight for dominance cant combine to form one coherent image. Occasionally in patched eyes the patched eye can win so world dark which is why you cant drive as yo cant control bin rivalry q

Question 50

Stereopsis and retinal disparity
How do we complete the viewing distances

Answer 50

Mag of disparity tells us how far away object is from horopter. Sign of disparity tells us if objects closer or beyond horopter.r

Vergence angles can be used as a depth cue. Closer object eyes inwards.= distance to horopter

Question 51

Basic strategy

Answer 51

The visual system uses a complex process to determine binocular disparities and perceive depth. The basic strategy involves identifying a visual feature on one retina, finding the corresponding feature on the other retina, and measuring the distance between these corresponding features to determine the object’s distance relative to the horopter.

However, this process is sophisticated because matching features between the two eyes can be challenging. This is evident in experiments with random dot stereograms, where patterns of randomly positioned dots are shown to each eye. Despite no recognizable patterns in the individual images, the shifted dots create a measurable disparity, causing the central square to appear in front of or behind the background, showing that disparity alone can create depth perception. This demonstrates the visual system’s ability to solve the correspondence problem, where it must correctly match dots between the two eyes out of many possible pairings. Despite the complexity, the visual system successfully fuses these images to perceive depth.

Question 52

Solving the correspondence problem

Answer 52

So if phoria then misalignment

To solve the correspondence problem, the visual system must correctly match images from both eyes to perceive depth accurately. In a simple scenario with two objects, A and B, the correct matches are images a with a’ and b with b’. Incorrect matches, such as b with a’ or a with b’, can create false perceptions of object positions. The visual system likely relies on binocular cells, which have slightly different receptive field positions in each eye and are tuned to specific disparities and distances. These cells are found in several visual areas, including V1, V2, V3, and MT.

Some individuals, known as stereoanomalous, can only use certain disparity information, suggesting different cell populations are tuned to crossed, uncrossed, and zero disparities. In the example, binocular cells with receptive fields at a in the left eye and a’ in the right eye correctly signal the object’s position, while cells at a in the left eye and b’ in the right eye incorrectly signal another position. All four cells in this setup will respond to the stimulus, leading to potential errors. This indicates that the visual system cannot rely solely on binocular cells for disparity measurement, as they might signal non-existent objects. Marr proposed a potential solution to address this issue.

Question 53

Computational theory

Answer 53

Computational Theory
In order to articulate his computational theory, Marr discussed the general problem in terms of random-dot stereograms. Marr (1982) outlined three important rules (constraints) that he saw as essential in solving the correspondence problem:

1) Compatibility. Black dots can match only black dots.
2) Uniqueness. Almost always, a black dot from one image can match no more than one black dot from the other image.
3) Continuity. The disparity of the matches varies smoothly almost everywhere over the image

Question 54

Algorithms

Answer 54

Marr and Poggio proposed a simple and biologically plausible solution to the correspondence problem. They suggested wiring binocular cells sensitive to similar dots together, embodying the idea that like-polarity dots match. This addresses rule 1. Additionally, since objects usually change smoothly in depth, cells signaling the same depth plane should support each other, explaining the excitatory connections between certain cells (rule 3). However, only one object can be seen along each line of sight from each eye, so cells signaling the same line of sight should inhibit each other, as described by rule 2.

Marr and Poggio built a computational network based on these principles, showing that it could solve the correspondence problem in complex random dot stereograms. Although the network, called the cooperative algorithm, takes several iterations to converge on a solution, it demonstrates how disparity-tuned neurons in biological vision could implement a similar mechanism.

Question 55

Stereo acuity

Answer 55

For objects placed at large viewing distances from the observer, stereopsis is of little or no use because the retinal disparities become too small.

However, at short viewing distances, depth judgements made from stereoscopic information are extremely keen. For example, at a viewing distance of 1m (1000 mm) it is possible to judge that one fine object is 0.5mm closer than another (this is a disparity of 0.0004 mm or 5 arcsec). For this reason, stereopsis turns out to have some rather interesting practical uses.

Question 56

Counterfeit bank notes

Answer 56

If bank notes had 0 disparities and everything same. One is diff bin disparities and shows up in BV and you see 3d relief. Camouflage. Busting used in second war.

Question 57

Iqras important points

Answer 57

Differences between 2 eyes= retinal disparities
Depend on relative object depth or distances
Stereopsis= recovering depth from retinal disparities
Magnitude of objects disparity= distance of object from horopter
Scaled according to distance from observer
Necessary for Stereopsis- same object seen at same time w each eye

Fixation point lies on vieth muller circle
Horopter= locus of points in physical world and 0/uncrossed disparity

If object is crossed disparity px needs to converge as object closer than horopter. Not a top down process cant recognise them
Disparity is inversely proportional to convergence so as convergence increases disparity decreases

2 times distance= distance decreases by a factor of 4
Disparity magnitude= dev from horopter/ D^2

Question 58

More important points

Answer 58

Correspondence problem= associating features across eyes
How can we solve this- random dot stereograms starts with 2 images of identical random dots and then central region displayed to left or right and gap filled with more random dots

Stereoblind observers- cant see any depth in stimuli

When images seen binocularly no depth is seen. Images look the same just dots.

Top down processing not importnat.
Mars rules- compatibility black dot matches black white matches white
Uniqueness= black dot forms 1 image and matches black dot from other
Continuity- disparity is smooth everywhere

Binocular rivalry when diff images presented to be
Images outside Panums= conflicting perceptions

Bin rivalry is a consequence of bv
Helpful to predators= can judge depths
Humans binocular overlap covers most of vf
Evidence for binocular mechanisms in humans
Interocular transfer of threshold elevation for contrast detection for sine wave gratings
Not evidence- discovery of disparity selective neurones in cortex

Question 59

Evidence for binocular mechanisms in human vision psychophysical

Answer 59

Interocular transfer of threshold elevation for contrast detection as shows combines inputs from both eyes to get binocular integration.

Or contrast detection for sine wave gradients. How adaptation in one eye affects perceived image in the other eye= integration.

Not evidence- neurophysical eg selective disparity neurons discovery or tilt aftereffect which can occur monocularly and the fact we have two eyes isnt specific enough. Ocular dominance columns discovery is neurophysiological. Some people stereo blind again doesnt show how evidence.

Evidence- Interocular transfer of tilt after effect, Interocular transfer of threshold elevation for contrast detection, Interocular transfer of the motion after effect. Interocular transfer of the sine change or sf after effect as all these things are adaptation effects key features of bv.

Question 60

Preys what do they have

Answer 60

Eyes on side of their heads so bv is helpful to them as it can provide a greater field of vision than achievable with a single eye. Amongst mammals there is less overlap in prey animals as sides of head so less bin overlap.

Predators have eyes facing forwards so greater binocular overlap enhancing their depth perception for head hunting.

Question 61

bin vision well developed

Answer 61

If it is well developed then there is a large mount of bin overlap and this results in teh computational problem of trying to achieve bv as large amount of bin overlap means the brain receives two slightly diff images and must combine them to make a single image

This involves compelx computational work which must be resolved to help and create single unified perception. In humans bin overlap covers most of hte visual field.

Question 62

In Stereopsis what produces zero disparity
What does the mag of an object disparity indicate

Answer 62

Only the object at the fixation point produces zero disparity

Mag of an objects disparity indicates the distance of the object from the horopter scaled according to the distance of the fixation point from the observer.

Question 63

What is the correspondence problem and how can it b solved

Answer 63

Correspondence problem is problem of associating a feature in one retinal image with the corresponding feature in the other retinal image.

Finding which parts of image seen. By one eye corresponds to which part of the image seen by the other eye. Needs to be solved to determine which points in the 2 retinal image corresponds to the same object in space.

It can be solved by Marr and pogios cooperative algorithm for stereo non transparent surfaces as they can match features between two retinal images. Non transparent surfaces= each point in image corresponds to 1 point in the other eyes image. Its challenge is matching the random dots to perceive depth.
They can be created by staring with two images of identical random dots and then displacing the central region of both images to either the right ot left by the same amount. 2 images with random dots- with slight horizontal displacement between corresponding points.

Purpose is to create depth so bin disparity can be perceived as depth. Cant be created with identical random dots need to displace and fill in gaps maybe with more random dots. Good for stereo blindness. Show that top down doesnt make contribution and its mainly bottom up where brain processes it.

Question 64

The horopter and Stereopsis

Answer 64

Only the object at the fixation point produces zero disparity.
Stereopsis needs same objects to be seen at same time w bothe yes

The horopter is the locus of points in the physical world beyond which disparity uncrossed. Or encloses crossed. Locsus of points in physical world that produces zero disparity in the eyes.

Locus of points in physical world that project onto corresponding points on teh retina- object or physical world to points on the retina.