Perception: Making

Meaning of Sensations
Naveen Kashyap, PhD
Indian Institute of Technology Guwahati
Email: naveen.kashyap@iitg.ernet.in
 Object Perception and Pattern Recognition
The process via which sensory inputs are gathered and meaningfully interpreted is called perception.
There are several forms of perception. When we look at an object we acquire specific bits of
information about it (location, kolor, shape, texture etc).

Is it true that we also at the same time when we perceive an object we also acquire information
about its function?

Classical approach to perception

Each of the objects/events in the real world like

trees, books etc are distal stimulus. The
reception and registration of information about
such o/e by sense organs make up the proximal
stimulus. The meaningful interpretation of the
proximal stimulus is the percept. Related to the
process of perception is a process called pattern
recognition – which is the recognition of a
particular o/e as belonging to a class of o/e.
Gestalt Approach to Perception
Gestalt approach interprets visual perception as how interpretation of stimulus arrays are done
as objects and backgrounds. For example the figure show two
distinct percepts: a white vase against a black background / two silhouetted
faces against a white background. This whole segregation of the whole
display into objects (figure) and the background (ground) is called form
perception. The part of the display seen as figure is seen to have a definite
shape and is better remembered than the ground which is less formed and
shapeless. Form perception happens as we assume intuitively that we
perceive objects and ground because they really are so and we merely
perceive them
The gestalt psychologists believed that perceivers follows certain laws or principles of organization
in coming to their interpretations The first assumed that the whole (gestalt) is not the sum of its part.

Gestalt principles of perceptual organization – there are

five major principles
a) Principle of proximity or nearness – this principle
allows us to group together things that are nearer to each
other.
b) Principle of similarity – allows us to group together objects that are similar

c) Principle of good continuation – states that we

group together objects whose contours form a
continuous straight or curved line

d) principle of closure – allows us to see images as complete by mentally

filling the gaps in the image

e) Principle of common fate – allows us to depict

elements that move together to be grouped together

Law of Pragnanz (Koffka, 1935) – is a general principle of gestalt that

houses all the other principle. It states that of all the possible ways of
interpreting a display, the organization that yields the most simplest and
stable shape or form is the one that is selected to interpret the element in
question.

Limitations of Gestalt approach

a) There is no explanation how these principles are translated into cognitive/physiological processes
b) The law of pragnanz seems to be circular (if additional specification is not given)
Bottom-Up Processes of Perception
The term bottom up (data driven) essentially means the perceiver starts with small bits of
information from the environment that he combines in various ways to form a percept.
example: a bottom up process of perception and pattern recognition might describe you seeing
edges, rectangular and other shapes and certain lighted regions and putting this information
together to “conclude” you are seeing doors and a hallway.

Picture = depth + figure + ground + texture + semicircular arch + semi circular block (door) like + ..
Template Matching – Every event, object or stimulus that we want to derive meaning from
is compared to some previously stored pattern or template. The process of perception in
template matching thus involves comparing incoming information to the templates we have
stored and looking for a match.
Limitations: a) requires a huge database to compare from
b) recognition of new objects
c) people recognize many patterns as more as less same thing
Featural Analysis – Instead of processing stimuli as whole units, we might instead break
them down into their components, using our recognition of those parts to infer what the
whole represents. The parts searched for and recognized are called features. Recognition of
a whole object in this model depends on recognition of its features.
Support for FA model: Studies done on retinas of frogs using microelectrode recording of single
cell revealed that certain stimulus caused these cells to fire more rapidly then certain others. Certain
cells responded strongly to borders between light and dark were called edge detectors, while certain
other cells responded selectively to moving edges were called bug detectors.

Irving Biederman’s (1987) theory of object perception – proposes that when people view
objects they segment them into simple geometric components like geons. Biederman proposed a
total of 36 such primitive components.
Biederman makes an analogy between his theory of object perception
with speech perception using phonemes (which are 44 in number and
are the basic unit of sound like /tS/ as in chair). As an evidence to this
theory Biederman offers the case of any fictional object that non of us
has seen but can try to decipher its parts with considerable agreement

Support for the featural theory also comes from the works of Eleanor Gibson (1969) which proved that
people are more likely to confuse with G & C then with G & F as G & C share the same features of
curved line open to the right.

Sefridge (1959) – Pandemonium model – consists of number of different kind of “demons”

which function basically as feature detectors. Demons at the bottom level (first) of processing scan
the input and demons at higher levels scan the output from the lower level demons. In response to
what they find, the demons scream.
Limitations of FAM – FAM suffers from the following shortcomings
a) There is no good definition of what can and cannot be a feature except the restricted domain
of perception of letter / line drawings
b) If there are different sets of features for different objects, how does the perceiver know
which ones to use to perceive an object

Prototype Matching – explains perception in terms of matching an input to a stored representation

of information as do template models. In this case, however the stored representations , instead of
being a whole pattern that must be matched exactly or closely, is rather a prototype – an idealized
representation of some class of objects or events – the letter M, a cup, a CAR etc.

According to prototype matching model – when a sensory device registers a new stimulus, the device
compares it with previously stored prototypes. An exact match is not required, only an approximate match
is expected. Prototype matching models allows for discrepancies between the input and the prototype. An
object is perceived when a match is found
Where do prototypes come from?
Posner & Keele (1968) demonstrated that people can form prototype very quick. They found that
people during an initial classification task, from some sort of mental representation of each class of
items.
Top-Down Processes
In top-down processing (also called theory-driven or conceptually driven processing) the
perceivers expectation, theories or concepts guide the selection and combination of the
information in the pattern recognition process. (for example)

You know from experience that archways generally mark alleys. When you look down the alley and
see it blocked in black you mostly expect a closed door etc………….
The context in which patterns or objects appear apparently sets up certain expectations in the perceiver
as to what objects will occur. Both accuracy and the length of time needed to recognize an object vary
with the context.
Top-down or conceptually driven processes are directed by expectations derived from context or
past learning or both.

David Marr – presented a computational and most elegant model of perception which involves
both the bottom up and top down process. According to this model visual perception proceeds by
constructing three different mental representations
a) primal sketch – depicts areas of relative brightness and darkness in a 2D images as well as
localized geometric structures. This helps in boundary detection
b) 2 ½ D sketch – using cues such as shading, texture edges and others the viewer derives
what the surfaces are and how they are positioned in depth relative to the viewers vantage
point
c) Final 3D sketch – involves both recognition of what the objects are and understand the
meaning of the visual scene
Perceptual Learning – perception changes with
practice has been well documented (E. J. Gibson, 1969),
and this phenomenon is called perceptual learning.
(Gibson’s original experiment with round coil cards).
Making individuals practice more with perceptual stimuli’s
enable them to learn what aspects of the stimulus to attend
to and try harder to consciously distinguish between
different kinds of stimuli. Using top-down processing the
perceivers experience guides him in selecting the most
optimal features to for more information
Change Blindness -
Change blindness – (Rensink, 2002) is the inability to detect changes to an object or scene, especially
when given different views of that object or scene and it illustrates the top-down nature of perception. The
change blindness paradigm reinforces the idea that perception is driven by expectations about meaning.
Instead of keeping track of every visual detail we instead seem to represent the overall meaning of the
scene.

The word superiority effect – word superiority effect or word

advantage advances that letters are apparently easier to perceive in
familiar context (a word) then in an unfamiliar or no context
environment.

Connectionist model of word perception – the model

assumes that input (written, spoken, thought) is processed at
several different levels, whether in terms of features, letters,
phonemes or words. Different levels of processing feed into
one another, with each level of processing forming a
representation of the information at a different level of
abstraction., with features considered less abstract then letters
and letters less abstract then words.

According to this model – perception of a word (activation of the

relevant node for the word) also activates the nodes corresponding
to all the letters within the word thereby facilitating their
perception
Direct Perception
Top-down and bottom down processes of perception believe that the perceiver does something to
the proximal stimulus for perception to proceed. This happens presumably because the proximal
stimulus doesn’t contain all the information we need to identify the object. This idea is called the
constructivist approach to perception. It describes people as adding to and distorting the information in
the proximal stimulus to obtain a percept.

James Gibson (1979) et.al., adopted an opposite view to the connectionist approach and believed
the perceiver does very little work in perception mainly because the word offers so much information
leaving little need to construction percepts and draw inferences. This view is called Direct Perception.
According to this view the light hitting the retina contains highly organized information that requires
little or no interpretation. In the world that we live in, certain aspects of stimuli remain invariant
despite changes over time or in our physical relationship to them.

Gibson became convinced that patterns of motion

provide a great deal of information to the perceiver. His
work with pilots in WWII led him to develop the idea of
optic flow as the visual array presented to a pilot
approaching the runway for landing. The arrows represent
perceived movement (apparent motion of clouds, grounds
etc wrt the pilot). There is a texture in the motion namely
nearer things appear to move faster and direction in which
objects seem to move depends on the angel of plane motion
in relation to them. These information are used by the pilot
to land the plane
 For Gibson the central question of perception is not how we look at and interpret the stimulus array
but how we see and navigate among real things in the world. An important ideal of Gibson’s theory is
that information available to an organism exists not merely in the environment but in an animal-
environment ecosystem.
Organisms directly perceive not only shape and whole objects but also each objects affordance’s –
the acts or behaviors permitted by objects, places and events (e.g. chair affords sitting, door knob
affords grasping, a glass window – looking etc)