Perception and Pattern Recognition

1
Chapter 3

• Perception and Pattern Recognition

2
Visual Perception
CHILVR Cornea, aqueus Humor, Iris, Lens,
Vitreous humor, Retina
3
Eye Anatomy

• Iris
• Cilliar Muscle
• Aqueous Humor
• Pupil
• Lens
• Cornea
• Retina
• Vitreous Humor

• Fovea
• Blind Spot
• Optic Nerve
• Rod
• Cone
• Ganglion Cell
• Bipolar Cell
• Horizontal Cell

4
The Fovea

• Light from direct focus lands in the Fovea.
• Contains almost all cones.
• Acuity -- accurate, precise vision-- is best in
  the fovea.
• Rods are abundant at the sides (periphery) of the
  fovea.

5
Sensation versus Perception

• Sensation
• The reception of energy from the environment, and
  its initial encoding into the nervous system.
• Perception
• The process of interpreting and understanding
  sensory information.

6
Sensation Perception
7
Gathering Visual Information

• Saccades Rapid eye movements.
• Last between 25 and 100 msec.
• Fixations Pauses between saccades.
• The eye takes in visual information during
  fixations.
• 200msec

8
Gathering Visual Information

• Change Blindness The failure to notice changes
  in visual stimuli (e.g. photographs) when those
  changes occur during a saccade.
• Inattentional Blindness We sometimes fail to
  see an object we are looking at directly, even a
  highly visible one, because our attention is
  directed elsewhere.

9
Visual Sensory Memory

• Also known as Iconic Memory.
• Visual persistence The apparent persistence of a
  visual stimulus beyond its physical duration (To
  demonstrate Rapidly wobble a pencil between two
  fingers).

10
Sensory Memory

• How do you study sensory memory?
• Because it is so brief it is difficult to study
• Sperling (1960) hypothesized that we have a
  visual memory
• He couldnt prove it because it always faded
  before people could report it
• What to do?

11
Sperlings Experiment (Overview)

• Used a tachistoscope to rapidly present images to
  the eyes.
• Subjects saw a 3 x 4 grid of letters, presented
  very briefly (50 ms).
• Subjects were asked to recall all the letters
  (Whole Report)
• Whole report performance was poor (about 37
  accuracy).

12
Whole Report

?

A N X B L F S M R P K V
(5-500ms)
(50ms)
13
Sperling (1960)

• In the whole report condition, subjects had to
  recall all the letters they could
• Performance was poor
• 4.5/12 letters
• 37 accuracy
• The span of immediate memory
• The number of individual items recalled after a
  short delay

14
Sperling (1960)

• But is this correct?
• All the letters might be available initially, but
  then fade before they can be reported
• If this is correct then subjects should be
  accurate reporting on any one row (before too
  much fading takes place)

15
Sperlings Experiment

• The partial report condition, in which only one
  of the rows was to be reported
• A tone was used as a cue
• A high pitch tone cued recall of the top row
• A medium pitch tone cued recall of the middle row
• A low pitched tone cued recall of the bottom row

16
Sperlings Experiment

• Partial report accuracy was much better
• 0 ms delay 76 accuracy
• So memory for entire display is about 76
• A lot more information is there, than can be
  reported
• 250ms delay 36 accuracy
• Like the whole report condition

17
Sperlings Experiment
18
Sperlings Experiment(Continued)

• In the partial report condition, a tone was
  sounded right after the letter grid disappeared.
• A high pitch tone cued recall of the top row a
  medium pitch tone cued recall of the middle row
  a low pitched tone cued recall of the bottom row.
• Partial report accuracy was at 76.

19
Sperlings Experiment(Conclusion)

• Why did Sperling argue that decay rather than
  interference is the loss mechanism in iconic
  memory?
• How long does information remain in iconic
  memory?

20
Lab Journal - Overview

• Only write up experiments on syllabus
• Single Page per Expt
• Number Pages -- everyones page 23 should have
  the same one.
• Use graphs when possible (picture 1000 words)
• Turn in at Midterm (for review) and then at Final
  Exam for scoring

21
(No Transcript)
22
One subjects results on the number of letters
available for report
23
Erasure and Masking Averbach and Coriell (1961)

• Presented two rows of eight letters each.
• Used a visual cue, above or below the
  to-be-recalled letter.
• The circle-maker cue masked perception of the
  letter.
• Backward Masking When a later visual stimulus
  affects perception of an earlier one.

24


X R F T L W Z P C A K N D E W J
50 ms
25


X R F T L W Z P C A K N D E W J
50 ms
26
Averbach Coriell (1961)

• With the bar cue accuracy was similar to
  Sperlings original findings
• High performance with short delays
• Lower performance with longer delays
• Effective duration of about 250 ms

27
Averbach Coriell (1961)

• With the circle cue performance decreased
  dramatically
• Accuracy was at chance
• Participants did not even see a letter in that
  location, they only saw the circle
• The circle erased or masked the memory for that
  letter

28
Averbach Coriell (1961)

• Backward Masking When a later visual stimulus
  affects perception of an earlier one
• Incoming information (circle) masked information
  that was already in iconic memory (letter)
• Implicit priming Information is still being
  processed, even though it has not been
  transferred into awareness

29
Ecological Validity and the Icon

• Haber (1983)
• The iconic memory is irrelevant to real-world
  perception.
• Data are useful only if reading in a lightning
  storm
• Criticisms of Haber?
• 1. Text reading task shows that perception
  actually is like reading in a lightning storm
  -- we only fixate for 50msec
• 2. Reductionism

30
Pattern Recognition

• How do we assemble visual input into meaningful
  shapes?
• Recognition of Letters
• Pandemonium
• Recognition of Objects
• Recognition by Components
• Context and Connectionism
• Visual Agnosia

31
Models of Pattern Recognition

• The Template approach
• Feature analysis / Feature detection
• Connectionism
• Recognition by components (object recognition)

32
Template Matching

• Categorize based on a Template (stored
  representation of all patterns)
• To perceive meaning, simply match the sensory
  input with a stored patten. When a match is
  found, you are done.

33
Template Matching

• Aa
• Aa
• Aa

34
Template Matching Problems

• Only possible in very limited domains (such as
  reading numbers 012345679)
• How would training work?
• Evidence clearly shows you can recognize (pattern
  match) objects youve never seen before.
• Conclusion -- probably not how the brain works,
  and probably not a great way to program a
  computer either.

35
Feature Analysis ? Letters

• Categorization by features
• A simple fragment or component that can appear in
  combination with other features
• Recognize letters by breaking them apart into
  features
• H A T L F

36
Feature Analysis ? Letters

• We recognize the features of letters (lines,
  curves) not the letters themselves
• Pandemonium Model (Selfridge, 1959)
• Decision Demons
• Cognitive Demons
• Computational Demons
• Data Demons

37
Pandemonium
38
Pandemonium Model(Selfridge, 1959)

• Demons Mental mechanisms that process visual
  stimuli ? Shout as they identify patterns
• Data Demons Encode a pattern
• Computational (Feature) Demons Feature
  analyzers each has a single feature it tries to
  match
• Cognitive Demons Each represents a letter of
  the alphabet wait for the right combination of
  features
• Decision Demon Has the final say in recognizing
  the pattern

39
Pandemonium Model

• Each node has only one simple job to do
• Includes parallel-processing
• Neurologically valid

40
Context Effects

• Data Driven (bottom-up)
• Processing is driven by the stimulus.
• Conceptually Driven (top-down)
• Processing is driven by higher level knowledge.

41
Examples of Top-Down Processing
42
Connectionism

• Input Units
• Hidden Units
• Output Units

• Back Propagation
• Delta Rule
• Distributed Representation
• Local Minima
• Massively Parallel Processing

43
(No Transcript)
44
Feature Analysis ? Objects

• Recognition by Components (RBC)
  (Biederman, 1987, 1990)
• All objects are a combination of geons ? Look up
  the combination to see which object matches
• Geons
• Feature detectors identify the geons based on

45
Geons The Objects They Make
46
Evidence for RBC

• Data on Non-Recoverable Drawings
• Impaired recognition if not able to identify the
  geons
• Degraded so no intersections
• Performance at chance
• When the same percentage of information was
  deleted but NOT at intersections (recoverable
  drawings) performance was at 75 accuracy

47
RBC

• Limitations of the model
• Only allows for data-driven processes (parsing
  and assembling)
• Whole vs. part perception
• Suggests that the whole is perceived by
  identifying the parts first ? but people perceive
  the overall shape as quickly and accurately as
  they perceive the components
• Neuropsychological evidence
• Object recognition is a joint effort between two
  mental processes ? one for features, one for
  global shape

48
Recognition by ComponentsBiederman (1987)

• Geons building blocks of visual objects.
• Data on Non-recoverable Drawings.
• Limitations of the model
• Over-reliance on data-driven processes?
• Whole versus part perception?
• Neuropsychological evidence?

49
Agnosia

• Failure or deficit in recognizing objects. Not
  blindness or poor vision.
• Apperceptive Agnosia (patterns)
• Associative Agnosia (meanings)
• Prosopagnosia Associative agnosia for faces

50
Example of Copying by a Patient with Apperceptive
Agnosia
51
Neuropsychology of Object Recognition

• Associative Agnosia
• Can copy drawings
• Can combine the features into a whole pattern,
  but cannot associate the pattern with meaning
• Can copy a drawing of an anchor and give an
  accurate definition of an anchor
• Cannot draw the anchor from memory nor identify
  the drawing that he copied
• Damage to both hemispheres (temporal lobe)

52
Example of Copying by a Patient with Associative
Agnosia
53
Neuropsychology of Object Recognition

• Prosopagnosia
• The Man Who Mistook His Wife for a Hat Ch1
• Associative agnosia that is specific for faces
• May not recognize self, or familiar friends and
  family
• Dissociation between face recognition and object
  recognition
• Are faces special?

54
Material not covered in class
55
Auditory Perception
56
Ear Anatomy

• Outer Ear
• Inner Ear
• Middle Ear
• Eardrum
• Ossicles
• Semicircular Canals
• Cochlea

• Oval Window
• Basilar Membrane
• Hair Cells
• Organ of Corti
• Round Window
• Auditory Nerve

57
Auditory Sensory Memory

• Also known as Echoic memory.
• Studying ASM
• Three-eared man technique
• Modality Effect
• Suffix Effect

58
Auditory Pattern Recognition

• Templates
• --the problem of invariance?
• Feature Detection
• Conceptually Driven Processing
• --Warren and Warrens (1990) study

59
Summary of Major Topics Covered in Chapter 3

• Visual Perception
• Pattern Recognition
• Object Recognition
• Agnosia
• Auditory Perception