PSY 416_ Pattern Perception_ 1

1 Pattern Recognition
How are you able to recognize things in certain patterns?
2

3 What is it?
■ What is pattern recognition? All the knowledge we use is based on this. A physician who was able to treat but recognize
diabetes. Everything we do involves pattern recognition.

■ Why is pattern recognition crucial?

Consider people who cannot recognize patterns Visual Agnosia, the inability to not recognize objects.
4 Case Described by Oliver Sacks:
Sacks “What is this?”
D.P. “May I examine it?”
A continuous surface, infolded in itself. It appears to have five
outpouchings, if this is a word”
Sacks: “Yes….tell me what it is” Contents, a glove. He had an infection in his frontal lobe, which
D.P. “A container of some sort” involves patterns. He was no longer able to tell things apart. Like his
Sacks: “Yes..and what would it contain?” wife and a hat.
D.P. : “It would contain its contents!!! There are many possibilities.
It could be a
change purse, for coins of 5 sizes. It could…”

5 Perception: Some Basic Concepts

■ Distal
Distal- obj that is out there in the distance, in the environment.
■ Proximal Proximal- obj that is close. On your retina and inverted and broken apart by blood vessels.
■ Retina Retinia- multilayer tissue in the back of the eye that contains rods and cones. They take visual info,
transduce it and make image. Cones are color perception. Babies aren’t able to see ccolor in the
■ Perceptual constanciesbeginning. Ther able to see red after black and white first. Babies have bad eye sight too, there basically
■ Size constancy blind at birth. Cones see detail in bright light. Rods light and dark perception. Function best at dim light.
■ Shape constancy Pupil expands in dim light to expose those rods.
Perceptual constancies, image on retina becomes bigger, the size is constant, the image changes, the
brain interprets it. Shape is consistent.
6

7 Gestalt Principles-Perceptual Organization

1. Gestalt: Whole may be greater than the sum of it’s part

Basically whole may be greater than the sum of its part. Breaking things into its elements is
worse. When you see the components all in one its much better then bresking it all apart. Its
much worse to do that.
Law of pragmaz- how we organize things w perception. These are describing how we do it.

1
PSY 416_ Pattern Perception_ 1

2. Law of Pragnanz: we tend to group elements that combine to form a good figure.

8 Proximity ( we see 2 colums rather then 4 rows, things grouped together are
perceived together), Similarity( columns of XOXO rather then rows, we group things
together how they are similar), Continuity(2 lines, one vertical and a curved one over,
9 rather then one over and over.), Closure (we like to close up the gaps)

10
3) Figure-Ground Relationship: Figure:

Walk down different environments, we are constantly seeing things as a figure, and
Ground:
then everything else is the ground. So background. It’s one over the other. Things pop
out as a figure.
11 Theories of Perception

12 A. Bottom-Up Theories

Data driven processing:

Information is out there, we combine information before we recognize it.
Data driven, begins w stimulus, we take it in and process it, then recognition occurs.
Stresses the importance of the stimulus in pattern recognition.
4 types
Template: In our brain we have all kinds of templates stored. We store it all in our
minds. And then recognize a pattern by matching it with one of our templates. Our
14 Bottom-Up Approaches visuals notice it. And it must be an exact match. You detect the info via rods and
[Link] cones, and then its transferred via nervous stimulus, and you observe it until you can
find a match, then our system recognizes it and stores it. There’s a lot of variation
[Link] within handwriting but were still able to recognize it. How. This model doesn’t work
[Link] well w human pattern perception. It does work well with matching a check w numbers.
[Link]-Description Theory Its very inflexible because u need a exact match. It might work w simple objects, w
more complex objects it starts to get harder and doesn’t work.

Prototype: have representation in ur head, and want to match something else in ur

15 Bottom-Up: Template Models head. Has to b similar enough. Think of a dog. That’s your protype. Info coming in has
■ We have stored in our minds sets ofto be similar enough to match w ur protype. That’s how u connect this dog, that dog.
templates.
Represent a mental average of examples, imagine 2 fam reunions at the park and u
have a good representation of each family, so a new person comes in and u try to
macth which fam this person belongs to bc of that. So u know what the fam looks like,
■ We recognize a pattern by comparing it with our set of templates.
u try to average that. U don’t need a exact match so this prototype theory is better. If
its not close enough, then the match doesn’t belong in this category.
■ We then choose the exact template that perfectly matches what we observe.
Feature: focuses on features. Distal stimulus out there, n instead of having figures in ur
mind u have features. You may have stored lines. So the features are looking for a
16 match. You have a feature detector, that looks for the features associated w what is
seen. Begins w stimulus entering sensory system. Research on this, feature analysis;
17 cats. Some features will fire when stimulus is only in very spechifc features. It is v
spechfic when they fire. That’s what electrodes told em. Like a line,only fired at
diagonal line, rather then straight line. In frogs there is a bug detectors, when
18 Template Theories
something crossess the visuals of a frog it sticks out there tounge to eat. Certain cells
detect certain features. (Neissser conducted research on this, used a stove, n within
19 Problems with Template Model miliseconds a person had to tell if they had found the particular feature. Can detect
■ Inflexible features when the background is v different. We do have a system that detects
features.

Structural description theory- built on feature detection theory and deals with 3d
objects, n components of them. Features and their relationships to one another w 3d
components. Recognization by component of the 3d object. Maybe 36 components,
and different enviroments are made up of these 36 different components. Like
everything is made up of the same stuff. Obj made up of the same geons but organized
in different ways to make objects. Language made up of phonems and we have like 40,
so we combine em and make different words w em. The geoms make up different
things. We recognize the components first and then weput them together to tell
objects. Geometric ions aka geoms.

2
PSY 416_ Pattern Perception_ 1

■ Works only for isolated letters and simples objects

20 Bottom Up: Prototype Models
More flexible versions of template matching theories.

21

22

23 Prototypes
■ Represent an averaging over a family of examples

■ More flexible than template theories

■ We compare a stimulus with a prototype

■ If a match is inadequate, we compare the stimulus with other prototypes

24 Bottom-Up Feature Theories
Feature Theories: We make discriminations between letters on the basis of a small number of
characteristics

27 Feature Analysis
■ Process of detecting specific elements and assembling them into more complex forms.

■ Brain cells “feature detectors” respond to specific

features, such as line and angles
■
■ Compatible with some biological research
■ Hubel & Wiesel’s research with cats
■ Lettiven et al.’s research with frogs

28
■ 1981 David Hubel and Torston Wiesel awarded Nobel Prize: Exploration of the mammalian
visual system.

3
PSY 416_ Pattern Perception_ 1

■ Some neurons function as “edge detectors” others as “dot detectors”

29 Other research related to featural processing:

■ Research using tachisoscopes
Visual Search Tasks (Neisser, 1963)

30

31

32
Was it easier to find the “M” in list A or list B?

Why?

33

34

36
4. Structural-Description Theory:

Built on feature detection theories

Includes feature & their relationship to each other.

37
Biederman
Recognition-By-Components:

a given view of an object can be represented as an arrangement of simple 3-D shapes called geons
(geometrical ions).

Essentially a feature theory for the recognition of 3-D objects

38

41

42 Summary
■ Bottom-Up Processing
Can have specific features that come into our system, and then recognition
43 occurs. So bottom up. Distal stimulus
Top down is conceptually driven, its what is all seen, perceiving the whole. So
bring ur knowledge to the situation and then see the whole picture and then
see the whole details. .