Professional Documents
Culture Documents
Chapter 4
Chapter 4
CHAPTER 4:
Cognitive
Psychology
American
University of
Beirut
Spring 2020
RECOGNIZING OBJECTS:
LECTURE OUTLINE
• Introduction: Visual agnosias
• The importance of features
• Experimental results in word recognition
• Models of word recognition (feature nets)
• Models of object recognition
• Face recognition
2
RECOGNIZING OBJECTS
3
VISUAL AGNOSIAS
4
INTEGRATIVE AGNOSIA
VISUAL AGNOSIA
6
RECOGNIZING OBJECTS
7
RECOGNIZING OBJECTS
8
RECOGNIZING OBJECTS
Incomplete
information
cat
9
RECOGNIZING OBJECTS
10
RECOGNIZING OBJECTS
Same stimulus
11
RECOGNIZING OBJECTS
Bottom-up processing
“Data-driven”, or stimulus-driven, effects
Feedforward neural projections?
Top-down processing
“Concept-driven”, or knowledge-driven, effects
Feedback neural projections?
12
TEMPLATE-MATCHING
13
TEMPLATE-MATCHING
Limitations
Variation in input (viewpoint, occlusion)
Infinity of templates
14
THE IMPORTANCE OF FEATURES
Features
Building blocks
Do not change much with viewpoint
Still present in occluded objects
Play a role in visual search
Visual search tasks: tasks in which participants are asked to examine a
display and to judge whether a particular target is present in the display
or not
Faster to look for one feature than to look for a combination of features
Feature detection is fast and occurs in parallel for many objects in the scene
15
THE IMPORTANCE OF FEATURES
17
VISUAL SEARCH
18
VISUAL SEARCH
19
VISUAL SEARCH
20
VISUAL SEARCH
T X T X
T T T T
X T X T T
X X X
T S T T X T
T T X T X T
X X
21
WORD RECOGNITION
22
WORD RECOGNITION
23
FACTORS INFLUENCING WORD
RECOGNITION
24
FACTORS INFLUENCING WORD
RECOGNITION
XJDKEL
TREE
TREE XJDKEL
Post-stimulus mask
Clears retinal image and sensory trace of stimulus
Interrupts continued processing
Factors influencing recognition time:
Word familiarity/frequency
Priming (repetition priming)
25
FACTORS INFLUENCING WORD RECOGNITION
26
FACTORS INFLUENCING WORD RECOGNITION
27
THE WORD SUPERIORITY EFFECT
28
THE WORD SUPERIORITY EFFECT
29
FACTORS INFLUENCING WORD/
LETTER RECOGNITION
Well-formedness
Strings of letters that resemble real words (pseudowords, e.g. HANE)
are recognized faster/better than those that don’t (e.g. random
strings: HNAE)
What is a well-formed string of letters?
Respects spelling regularities of the language (organized like existing
words)
Frequency of letter combinations
Well-formed strings also produce a word “superiority” effect
compared to random letter strings
Could suggest that words are perceived as a whole as fast as, or
faster than, individual letters, and in parallel with them
30
MAKING ERRORS
(OVER-REGULARIZATION)
Well-formedness also influences errors
Example of a likely error: “DPUM” misread as “DRUM”
Example of an unlikely error: “DRUM” misread as “DPUM”
Reading quickly-presented words that also follow common spelling
patterns
Non-words or pseudowords mistaken for real words
People can perceive stimuli as being more “regular” than they
actually are
Example: “TPUM” misread as “TRUMPET”
31
FEATURE NETS AND WORD RECOGNITION
32
FEATURE NETS AND WORD RECOGNITION
Complex
Simple
33
FEATURE NETS AND WORD RECOGNITION
34
FEATURE NETS
36
FEATURE NETS AND WELL-FORMEDNESS
38
RECOVERY FROM CONFUSION
39
RECOVERY FROM CONFUSION
40
RECOGNITION ERRORS
41
AMBIGUOUS INPUT
45
DESCENDANTS OF THE FEATURE NET
46
THE MCCLELLAND AND RUMELHART MODEL
INTERACTIVE ACTIVATION MODEL
This is a much more complex feature net, with feedforward,
lateral, and feedback loops
Higher level word detectors can influence lower level detectors
Information flows bottom-up, top-down, and within the same level
Detectors at any level can also influence other detectors at the same
level (e.g., letter detectors can inhibit other letter detectors; word
detectors can inhibit other word detectors)
There are both excitatory connections and inhibitory connections
Excitatory connections allow one detector to activate its neighbors
Inhibitory connections allow detectors to inhibit their neighbors
Bigram detectors are not necessary in this model
Word superiority and well-formedness effect explained by feedback
connections between word letter detectors
47
THE MCCLELLAND AND RUMELHART MODEL
INTERACTIVE ACTIVATION MODEL
The McClelland and Rumelhart (1981) pattern-recognition model includes
both excitatory connections (indicated by red arrows) and inhibitory
connections (indicated by connections with dots)
Connections within a specific level are also possible
For example, activation of the “TRIP” detector will inhibit the detectors for “TRAP,”
“TAKE,” and “TIME”
48
THE MCCLELLAND AND RUMELHART MODEL
INTERACTIVE ACTIVATION MODEL
In original model, feature detectors can only increase
activation of other features
In this model, some detectors can also inhibit other detectors
That is, decrease the other detector’s activation when they are
activated
Introducing inhibitory and top-down connections can explain
well-formedness and word superiority effects without need for
bigram detectors
Word detector activated by certain letters will inhibit letters that do
not appear in the word
Overall, letters compatible with words that contain detected letters
will be activated more than letters that are not compatible with these
words
Existence of feedback and lateral connections correspond well
to what we know of the brain
49
INTERACTIVE ACTIVATION MODEL
50
RECOGNITION BY COMPONENTS
51
RECOGNITION BY COMPONENTS
52
RECOGNITION BY COMPONENTS
Hierarchy of detectors
Feature detectors (e.g. curves,
Object model
edges)
Geon detectors
“Geon assemblies” Geon assemblies
representing relations
between geons
Object model: a 3D
representation of complete,
recognized object
Geon detectors
Bottom-up recognition
Feature detectors
53
RECOGNITION BY COMPONENTS
54
RECOGNITION BY COMPONENTS
Different geons
55
Prime Target
RECOGNITION BY COMPONENTS
Viewpoint-independent
Geons can be identified from virtually any angle of view
Thus, the object model can be activated and the object can be
recognized
Example: No matter what your position is relative to a cat, you’ll be able
to identify its geons and thus identify the cat (even from just a few geons)
Partial occlusion of objects does not necessarily prevent recognition
Moreover, it seems that most objects can be recognized from
just a few geons
As a consequence, geon-based models like RBC can recognize
an object even if many of the object’s geons are hidden from
view
RBC model limited to recognition of objects with different 3D
structures, not fine discrimination of objects with the same
structure (e.g. face recognition)
56
RECOGNITION VIA MULTIPLE VIEWS
57
RECOGNITION VIA MULTIPLE VIEWS
Hierarchy of detectors
Each successive layer processes more complex aspects of the whole
Detectors represent a particular viewpoint
58
RECOGNITION VIA MULTIPLE VIEWS
59
RECOGNITION VIA MULTIPLE VIEWS
60
RECOGNITION VIA MULTIPLE VIEWS
61
RECOGNITION VIA MULTIPLE VIEWS
62
RECOGNITION VIA MULTIPLE VIEWS
63
RECOGNITION VIA MULTIPLE VIEWS
64
FACE RECOGNITION
S om e pe o ple a r e “ s u per - r e c og ni ze rs ”
They are extremely accurate at face recognition even faces viewed only briefly at some
distant point in the past
More successful at tasks that require “face matching”
Judging whether two different views of a face actually show the same person
But they have no other perceptual- or memory-based advantages
66
FACE RECOGNITION:
FACE INVERSION EFFECT
67
FACE RECOGNITION
68
FACE RECOGNITION
69
FACE RECOGNITION:
THATCHERIZED FACES
Is there anything wrong with these faces?
70
FACE RECOGNITION:
THATCHERIZED FACES
How about now?
71
FACE RECOGNITION
72
HOLISTIC RECOGNITION
73
HOLISTIC RECOGNITION
Who is this?
74
HOLISTIC RECOGNITION
Who is this?
75
HOLISTIC RECOGNITION
Composite effect
The relationship among features guide face recognition
76
HOLISTIC RECOGNITION
77
HOLISTIC RECOGNITION
78
HOLISTIC RECOGNITION
Composite effect:
Comparison of features is easier in composite than in whole faces
Not observed for (most) other objects
The relationships among features guide face recognition
79
HOLISTIC RECOGNITION
80
HOLISTIC RECOGNITION
81
TOP-DOWN INFLUENCES ON
OBJECT RECOGNITION
Limits of feature nets:
Some target objects depend on configurations, not individual
features
Knowledge that is external to object recognition nevertheless
influences recognition
For example, words are easier to recognize as part of a sentence than in
isolation
Semantic priming aids recognition
If someone tells you you’re going to see the name of a food item, you’re much more likely to
identify the word CELERY
In this case, word recognition has been influenced not by visual properties of the
words (or their frequency/familiarity), but by their meaning
Object recognition is not a self-contained process, but can be
influenced by other cognitive processes such as memory, knowledge,
etc… (beyond word familiarity effects)
82
TOP-DOWN INFLUENCES ON
OBJECT RECOGNITION
W he n v ie w e r s h a d o n l y a v e ry b rie f g lim p se o f a t a rg e t o b je c t , b ra in a c t iv it y
i nd ic a t ing t o p - d o w n p r o c e ssin g wa s e v id e n t in fr o nt p a rt o f b ra in (o rb it o f ro nt a l
c o r t e x ) 1 3 0 m s a ft e r t a r g e t c a m e int o v ie w
R o u g hly 5 0 m s l a t e r (a n d t hu s 1 8 0 m s a f t e r t a r g e t c a m e int o v ie w), b ra in
a c t iv it y in c r e a se d f u r t h e r b a c k in b ra in (i n t h e rig ht he m isp h e re ’ s f u sif o rm
a re a ), ind i c a t i n g s u c c e s sf u l r e c o gn it io n
T his p a t t e r n n o t e v i d e n t w h e n o b je c t re c o g nit io n w a s e a sy (b e c a u se o f lo n g e r
p re se nt a t i o n o f t a r g e t )
S e nsib ly , t o p - d o w n p r o c e ssin g p la y s la rg e r ro le whe n b o t t o m - u p p ro c e ss ing is
s o m e ho w l i m i t e d o r i n a d e q u a t e
83
CHAPTER 4 QUESTIONS
QUESTION 1
85
QUESTION 1: ANSWER
86
QUESTION 2
87
QUESTION 2: ANSWER
88
QUESTION 3
a. Viewpoint-dependent
b. Viewpoint-independent
c. Object
d. Face
89
QUESTION 3: ANSWER
a. Viewpoint-dependent
b. Viewpoint-independent
c. Object
d. Face
90
QUESTION 4
91
QUESTION 4: ANSWER
92
QUESTION 5
a. Apperceptive agnosia
b. Integrative agnosia
c. Prosopagnosia
d. Associative agnosia
93
QUESTION 5: ANSWER
a. Apperceptive agnosia
b. Integrative agnosia
c. Prosopagnosia
d. Associative agnosia
94
QUESTION 6
95
QUESTION 6: ANSWER
96
QUESTION 7
97
QUESTION 7: ANSWER
98