Perception

 The Nature of Perception

o Perception is defined as experiences resulting from stimulation of the senses

o Some basic characteristics of perception:

 Perception can change based on new data such as moving closer or

farther from an item can shift perception

 It can involve problem-solving or reasoning such as trying to figure out

what a distant object is

 It can be based on a perceptual rule (Gestalt principles)

 It is not an automatic process (as automatic processing refers to a type

of cognition that does not involve any effort) and it can occur in parallel

with action as we are constantly in motion

 In fact, perception is central to one’s ability to organize the actions that

occur as one interacts with the environment; it is also important for

cognition as it is essential for creating memories, problem-solving,

communicating with others etc

 A Computer-Vision System Perceives objects and a scene

o It is important to note that human perception comes naturally and so one may

think it is a meh element however one way to appreciate its complexities is to

consider how difficult it is for software programs to accomplish perceptual tasks

that humans achieve with ease

o Designing a machine vision system that can perceive and interact with its

environment is not easy as Star Wars makes it seem (think C-3PO). These

systems which are the basis for driverless cars were not accomplished until
 quite recently and now they are quite commonly found in the Western World

and yet these machines still make quite a number of mistakes

o Researchers have observed that computer-vision systems still have difficulty

naming objects, so while technology has crossed many strides, creating a

computer-vision system such as C-3PO is not yet a complete goal though several

AI companies are racing to achieve it

 Why is it so Difficult to Design a Perceiving Machine?

o Stimulus on the receptor is ambiguous:

 When we look at an object, the image that falls on the retina is inverted

and the perceptual system starts with this retinal image as its job is to

identify which object created this image. This task of determining which

object is known as the inverse projection problem which can be

illustrated with rays starting at the retinal image and extending

outwards

 By tracing these rays, we can observe that the same retinal image could

have been caused by different objects which makes the retinal image

ambiguous, human brains solve the inverse problem quite easily but still

poses some difficulty for computer-vision systems

 Objects Can Be Hidden or Blurred

o It might take you a couple of seconds to locate the hidden pencil and glasses but

eventually you will find them

 This issue of hidden objects is frequent occurrence in the environment

and humans are quite good at identifying objects even if they are

partially hidden and that is thanks to object permanence (which refers

to the understanding that even if an object is hidden it is still these) an

one’s knowledge

o Humans are also able to identify blurry objects or faces whereas computers face

difficulty with this as well as locating hidden objects

o Another problem computer-vision systems face is viewpoint invariance (which

refers to the ability to recognize an object even when it’s seen from different

viewpoints)

 These systems can achieve this only through a long process of complex

calculation designed to determine which points on an object match in

different angles

o When it comes to perceiving scenes, these are much more complex than a

single object as they hold more data which some would require reasoning based

on experience. This kind of data would have to be programmed into a

computer, step by step, for it to be able to perceive a scene and figure it out as

quickly as a human brain can

 Information for Human Perception

o Perception is built on a foundation of information from the environment and t

relies on 2 types of processing;

 o Bottom up processing:

 Refers to processing that starts with information received by sensory

receptors and goes all the way to the CNS (which includes the brain and

spinal cord)

o Top-down processing:

 Refers to processing that involves a person’s knowledge or expectations

as it starts at the level of the brain and travels down to the rest of the

nervous system

 Hearing Words in a Sentence

o An example of top-down processing influencing speech perception could be

illustrated through listening to people conversing in a foreign language

 We can perceive that their speech is continuous and identify individual

words from one another even without understanding the meaning,

though of course knowledge of a language facilitates this

o This ability to determine where one word starts and another ends is identified

as speech segmentation

 basically 2 people with different linguistic background (one knowns

French and the other Spanish for example) would perceive a similar

speech stimulus differently; sound entering the ear and sending signals

to the brain is common to both (bottom-up processing) but if the

listener knows the language then they are able to perceive and

understand individual words (top-down processing)

o while knowledge of language certainly facilitates speech segmentation, there

are other information one can use to achieve this as one is in the process of

learning a language

 basically, when learning a language, we are also learning about its

transitional probabilities, which refers to the likelihood that one sound

will follow another within a word. So in English for instance, if we were

to consider the words cheesy sandwich, you know that it is more likely

for “chee” to be followed by “sy” than “sand”

o there are other characteristics of language that once picks up while learning a

language and these are catalogues under the concept of statistical learning

which refers to the process of learning about transitional probabilities and

about other characteristics of language (it also occurs for vision, based on

learning about what types of things usually occur in the environment

o research shows that children as young as 8 months old are capable of statistical

learning

o Saffran et al. (1996) carried out a learning design experiment on young infants

who would first undergo a learning phase of 2 minutes of listening to 4

nonsense words put in a continuous string which is to make it difficult to tell one

word from another

o Generally, babies get bored quickly with familiar stimuli and so it was

hypothesized that if statistical learning occurs, then when exposed to a series of

individual nonsense words including the 4 learnt, infant are more likely to listen

longer to new words as opposed to those that were learnt

 o Results support this which in turn confirms the importance of top-down

processing in perception

 Conceptions of Object Perception

o Helmholtz’s theory of unconscious inference:

 Hermann von Helmholtz, a beautiful mind, was a physicist who made

several important contributions to thermodynamics, nerve physiology,

vision, psychology and few other fields. He also invented the

ophthalmoscope to examine the eye

 When it comes to perception, Helmholtz was one of the 1 st to observe

that the image on the retina is ambiguous (ambiguity refers to how a

particular pattern of stimulation on the retina can be triggered by

several objects)

 He sought to examine how does the perceptual system decide that a

specific pattern on the retina can be created by a certain object. To

answer this, he developed the likelihood principle which states that we

perceive the object that is most likely to have triggered the pattern of

activation in the retina and this is determined through a process known

as unconscious inference in which our perceptions are the result of

unconscious assumptions/inferences we make about the world. This

process is similar to problem-solving because the perceptual system

would have to rely on one’s knowledge to infer what objects in the

environment are
  It is important to note that this is a fast, unconscious, almost automatic

process that one goes through when trying to figure out a scene for

instance or when trying to locate a friend in a crowd

 The Gestalt Principle of Organization

o 30 years after Helmholtz’s work, Gestalt psychologists took up a different

approach to understanding perception

o Gestalt psychologists rejected the notion that perception was formed by adding

up sensations as structuralists would put it; and this was achieved through the

concept of apparent movement (refers to when motion is perceived even when

there isn’t any) which is observed by Max Wetheimer, on a train while looking at

a stroboscope (a device that creates the illusion of movement by alternating 2

slightly different pictures very fast), and this led him to the following conclusion:

 Apparent movement can’t be explained by sensation because as the

lights turn on and off, when there’s darkness between the flashing lights

it means there is no sensory input here

 The second conclusion, later became a main principle of Gestalt

psychology which is that the whole is different than the sum of its parts,

as one’s perceptual system is capable of creating the perception of

motion from still images

 This concept led to the proposal of several principles of

perceptual organization which are basically rules to explain how

small elements of a scene or display become perceptually

grouped to form larger units

 o First we have the principle of good continuation which states that 2 points, that

when connected result in a straight or smoothly curving line, are seen as

belonging together, and the lines tend to be seen in such a way as to follow the

smoothest path. It also states that objects are covered by another object are

perceived as continuing behind the overlapping object

o The law of prognanz or principle of good figure or principle of simplicity which

states that every stimuli pattern is seen in such a way that resulting structure is

a simple as possible

o The principle of similarity which states that similar items appear to be grouped

together

o There are4 more principles but what’s important for us to note is that Gestalt

psychologists realized that perception is determined by specific organization

principles which are described as “intrinsic laws”

o These principles are consistent with Gestalt’s idea that a person’s experience

play only a minor role in influencing perception, which is contradictory to

Helmholtz’s likelihood principle which puts knowledge at the centre

 Taking Regularities of the Environment in to Account

o Modern perceptual researchers take experience into account by identifying the

frequently occurring characteristics of the environment which are referred to as

regularities in the environment and there are 2 types to explore, physical and

semantic regularities

o Physical regularities refer to regularly occurring physical properties of the

environment.
  An example of this would be how the world is made up of more vertical

and horizontal lines as opposed to oblique line hence why we can

perceive horizontal and vertical lines much easier than oblique ones or

any other orientation, this is known as the oblique effect

 Another example would be the light from above assumption which is

that we assume light be it natural or artificial is coming from above

 It is important to note that the reason humans are better at detecting

physical regularities than computer-vision system is because we have

adapted to the regularities of the world

o Semantic regularities, when it comes to perception, refer to the characteristics

associated with the functions carried out in different types of scenes.

 This can be portrayed in picturing a restaurant, generally one would

hold in mind certain characteristics in mind of what a restaurant should

look like to be called as such. This knowledge of what a given scene

contains is identified as scene schema and it generates expectations of

what something should look like.

o This use of regularities to perceive complex scenes within the natural world is

not a conscious process

 Bayesian Inference

o In the 1700s, Thomas Bayes proposed that one’s estimate of the probability of

outcome is determined by the prior probability (initial belief about the

probability of an outcome) and the extent to which the available evidence is

consistent with the outcome, i.e. the likelihood outcome

 o In practice this involves a mathematical procedure but the basic idea is that

people start with prior which can be updated with additional evidence to reach

a conclusion

o It is important to note that the Bayesian inference, Helmholtz’s unconscious

inference and the concept of regularities are complimentary as they assert the

importance of top-down processing and experience in perception. The approach

by Gestalt psychologists adds to this by stating that in addition to experience

these principles of organization also have a role to play, consider them as hacks

for your perceptual system

 Neurons and Knowledge about the Environment

o Neurons that respond to horizontal and verticals:

 Recall the oblique effect, so not only are we more receptive to

horizontal and vertical lines but also hold more neurons who are

receptive to these 2 orientations over any other orientation

 A possible explanation of this could be based on the theory of natural

selection which states that characteristics that enhance an animal’s

ability to survive and reproduce will be passed onto future generation

 Since horizontals and verticals are predominant in nature, our

brains are more likely to hold neurons that respond to these

lines more than neurons responding to other orientations

 Evolution has of course shaped perceptual functioning but it is not the

only force at play here as learning can shape a neuron’s response

properties through experience-dependent plasticity

 Experience-Dependent Plasticity
 o Recall Blakemore and Cooper’s experiment on selective rearing with cats in

which the cats were placed in either horizontally or vertically aligned

environment and as a result the cats were found to have neurons only for the

orientation but they were reared in to

 This provides evidence that experience can shape the nervous system

hence the term experience-dependent plasticity

o Conclusion: experience-dependent plasticity shows that the brain’s functioning

can be adapted to its environment

 Perception and Action: Behavior

o Movement facilitates perception:

 Movement allows us to see different aspects of an object especially for

objects that are not very clear from one viewpoint and so movement

would help view it from different viewpoints

o The interaction of perception and action:

 Movement is also important as one is continuously engaging in some

form or another with their environment and so there is a continuous

coordination between perceiving stimuli and taking action towards

them

 Perception and Action: Physiology

o Physiology brought some clarity to understanding this connection between

perception and action, specifically this came from physiological research in the

1980s which showed that there are 2 processing streams

 o There were 2 mains methods applied by researchers to explore this, either brain

ablation, which refers to the study of the effect of removing certain brain areas

in animals, and neuropsychology

 What and Where Streams

o In 1982, researchers Ungerleider and Mishkin performed, what can be termed

today as a classic experiment, a brain ablation on monkeys to see how it would

affect their performance on object recognition and localization. The monkeys

were presented with 2 tasks to observe if the ablation had any effect:

o This object discrimination problem (object recognition task)

 Monkeys here had parts of their temporal lobe removed and were

asked to identify an object that they had seen before.

 Performance was affected by the missing part of the temporal lobe

which became known as the seat of the “what” stream/pathway (striate

cortex to the temporal lobe)

 Ventral stream

o The landmark discrimination problem

 The task here being to remember an object’s location and select it after

a delay; it involved having parts of the parietal lobe removed.

 Performance was affected and this area became associated with the

“where” pathway which goes from the striate cortex to the parietal lobe

 Dorsal stream

 Perception and Action Streams

o Milner and Goodale (1995) studied the case of D.F., a woman whose temporal

lobe was damaged and based on their results or observations throughout

 certain tasks such as asking D.F. to match a card to its orientation or to mail a

card.

 They suggested that the stream of information going from the visual

cortex to the temporal lobe be identified as the perception pathway

(which would correspond to the what pathway) as she performed

poorly when no action was involved

o However, when asked to mail the item into the slot, she was able to push it

through without difficulty and this involved the pathway from the visual cortex

to the parietal lobe.

o This stream became known as the action or how pathway (also known as the

where pathway)

 Mirror Neurons

o Researchers studying monkeys, throughout a running experiment, observed

that when the researcher dropped food, the monkey observing this action or

rather more specifically their premotor cx showed an activity pattern as if it was

the monkey who was picking up the food

o This observation along with other experiments led to the discovery of mirror

neurons and these refer to neurons that respond both when performing and

observing an action

o Studies exploring this in humans have observed that mirror neurons are spread

out in the cortex as part of a network identified as the mirror neuron system

o The purpose of this system is still an active debate but the leading theory thus

far suggests that it can allow one to determine the intention behind an action.