1 perceptual processes introduction pattern recognition pattern recognition top-down processing...
TRANSCRIPT
1
Perceptual Processes
Introduction Pattern Recognition Top-down Processing & Pattern Recognition Face Perception
Attention Divided attention Selective attention Theories of attention
2
Perception
Process that uses our previous knowledge to gather and interpret the stimuli that our senses register
7
Theories of Pattern Recognition
Template Matching Theory
Prototype Models
Distinctive Features Model
Recognition by Components Model
8
Template Matching Theory
Compare a new stimulus (e.g. ‘T’ or ‘5’) to a set of specific patterns stored in memory
Stored pattern most closely matching stimulus identifies it.
To work – must be single match
Used in machine recognition
11
Problems for Template Matching
Inefficient - large # of stored patterns
required
Extremely inflexible
Works only for isolated letters and simple
objects
12
Prototype Theories
Store abstract, idealized patterns (or
prototypes) in memory
Summary - some aspects of stimulus
stored but not others
Matches need not be exact
15
Prototypes
Family resemblances (e.g. birds, faces,
etc.)
Evidence supporting prototypes
Problems - Vague; not a well-specified theory of pattern recognition
16
Distinctive Features Models
Comparison of stimulus features to a stored list of features
Distinctive features differentiate one pattern from another
Can discriminate stimuli on the basis of a small # of characteristics – features
Assumption: feature identification possible
17
Distinctive Features Models: Evidence
Consistent with physiological research
Psychological Evidence Gibson 1969 Neisser 1964 Waltz 1975 Pritchard 1961
Can we empirically test the distinctive features theory?
• In other words, can we show that we must be processing features when we identify and distinguish one pattern from another – e.g. letters?
• There are many ways we can test a feature-based theory.
• For example:
20
21
Scan for the letter ‘Z’ in the first column of letter strings.
Scan for the letter ‘Z’ in the second column of letter strings.
Where did you find the ‘Z’ faster: in column 1 or 2? What does this show?
25
Distinctive Features
Theory must specify how the features are combined/joined
These models deal most easily with fairly simple stimuli -- e.g. letters
Shapes in nature more complex -- e.g. dog, human, car, telephone, etc
What would the features here be?
26
Recognition by Components Model
Irving Biederman (1987, 1990) Given view of object can be represented
as arrangement of basic 3-D shapes (geons)
Geons = derived features or higher level features
In general 3 geons usually sufficient to identify an object
28
Status of Recognition by Components Theory
Distinctive features theory for 3-D object
recognition
Some research consistent with the model;
some not
Recognition by Components Pro – Biederman found that obscuring vertices impairs object recognition while
obscuring other parts of objects has a lesser effect.
29
Which is easiest to recognize as a cup? The left or right?
Con – Biederman – Not all natural objects can be decomposed into geons. What about a shoe?
31
Summary
Distinctive Features approach currently strongest theory
Perhaps all 3 approaches (distinctive features, prototypes, recognition by components) are correct
Regardless, pattern recognition is too rapid and efficient to be completely explained by these models
32
Two types of Processing
Bottom-up or data-driven processing
Top-down or conceptually driven
processing
Theme 5 -- most tasks involve bottom-up
and top-down processing
33
Thought Experiment
Assume each letter 5 feature detections involved
Page of text approximately 250-300 words of 5
letters per word on average
Each page: 5 x 5 x 250-300 = 6250 - 7500
feature detections
Typical reader 250 words/min reading
6250/60 secs =100 feature detections per
second
38
Word Superiority Effect
We can identify a single letter more rapidly and more accurately when it appears in a word than when it appears in a non-word.
44
Altered Sentences in Warren and Warren (1970)
Sentence that was presented Word Heard
It was found that the *eel was on the axle
It was found that the *eel was on the shoe
It was found that the *eel was on the orange
It was found that the *eel was on the table
wheel
heel
peel
meal
*Denotes the replaced sound
49
Reinitz & Colleagues (1974)
Divided Attention Condition
Subjects count the dots
Full Attention Condition
No instruction about dots
50
Proportion of Responses that were “old” for Each of Two Study Conditions and Two Test Conditions
(Reinitz & Colleagues, 1994).
Study Condition
TestCondition
Full Attention Divided Attention
Old Face
ConjunctionFaces
.81
.48
.48
.42
Cell Phones & Driving – What Does the Research on Divided Attention Show?
• Is it safe to drive while talking on a cell phone?• Some states have passed legislation prohibiting
hand-held but not hands-free cell phones. Does this make any sense
• What about talking to someone in the car while driving versus talking on a cell phone?
• What are the chances of an accident?• Does practice make a difference? Why not?• Compare driving under the influence to cell phone
driving55
57
Selective Attention (Dichotic Listening Task)
Shadowing
Irrelevant Channel
Cocktail Party Effect - Morray (1959)
Wood and Cowan (1995)
Treisman (1960)
Read the Word.
68
Green
Blue
Red
Purple
Green
Blue
Orange
Red
Blue
Green
Red
Blue
Orange
Green
Blue
Red
Purple
Orange
Green
Black
Stop!
Read the Word. Ignore the color
69
Green
Blue
Red
Purple
Green
Blue
Orange
Red
Blue
Green
Red
Blue
Orange
Green
Blue
Red
Purple
Orange
Green
Black
Stop!
Name the Color of the Ink
71
xxxxx
xxx
xxx
xxxxxx
xxxxx
xxxxx
xxxxxx
xxx
xxxx
xxxxx
xxx
xxxx
xxxxxx
xxxxx
xxxx
xxx
xxxxxx
xxxxxx
xxxxx
xxxxx
Stop!
Name the Color (e.g. Red say “blue”)
72
Green
Blue
Red
Purple
Green
Blue
Orange
Red
Blue
Green
Red
Blue
Orange
Green
Blue
Red
Purple
Orange
Green
Black
Stop!
Stroop’s 3 Experiments• Exp 1 - Selectively attend to the verbal
aspect of the stimulus; ignore ink color
• Exp 2 – Selectively attend to the ink color of the stimulus; ignore verbal aspect
• Exp 3 – Why does ignoring the verbal aspect of the stimulus interfere strongly with color naming; but not the reverse?
73