Visual Attention

advertisement
Attention
Looking without Seeing
Why Have Attention?
• Limited resources
– Too much information
• Attention:
1. selects important/relevant information
2. modulates it in the context of the task at hand
Attention Mechanisms
• Top-Down
– Goal-driven
• Bottom-up
– Stimulus-driven (“attention capture”)
– There is debate if total bottom-up really exists
• Attention capture is shown to be modulated by task
goals
• Early Selection vs. Late Selection
Visual Attention
15.1
Evidence for Early Selection in Audition
Shadowing paradigm
What can be followed?
+
+
X
Position (left/right)?
Pitch (male/female)?
Language (English/French)?
--> Early Selection
© Stephen E. Palmer, 2002
15.1
Visual Attention
Evidence for Late Selection in Audition
Cocktail Party phenomenon
Unselected information can get in:
Subject’s own name
Words expected from context
--> Early and late selection
© Stephen E. Palmer, 2002
15.1
Visual Attention
Inattention Paradigm (Mack & Rock)
What do we see without attention?
D. Divided Attention
Trial
© Stephen E. Palmer, 2002
15.1
Visual Attention
Inattention Paradigm Results:
P e rform anc e re lative to C han ce
.
Perfect
Performance
100
Location
80
Color
60
40
Number
Shape
20
Chance
Performance
0
Inattention
Divided
Control
Trial Condition
© Stephen E. Palmer, 2002
Visual Attention
15.1
Inattentional Blindness:
On many trials, subjects
report seeing NOTHING if
the test object is at fixation.
Square at Fixation: 50-75% IB
Own Name at Fixation: 5% IB
Other’s name at Fixation: 35% IB
Variant of own name at Fixation: 60% IB
(e.g., JECK instead of JACK)
© Stephen E. Palmer, 2002
Show movie…
Visual Attention
15.1
Change Blindness
© Stephen E. Palmer, 2002
Visual Attention
15.1
Change Blindness
© Stephen E. Palmer, 2002
Eye Movements
Q: Why?
A: Limitations of the eye – only fovea is high-res
enough for many tasks
Two types:
– Saccades
• Rapid motion (25-30 ms) between fixations
• Saccades occur every 250-300 ms
• Also evidence for “micro-saccades”
– Smooth-pursuit (tracking) movements
• Require feedback
Eye-Tracking
Alfred Yarbus
goal-attenuated
Alfred Yarbus
Saliency Maps
• Itti et al proposed that bottom-up attention
can be predicted from low-level visual
features.
• Eye-tracking can be used to validate the
predictions
• What are the problems with this idea?
Download