Perceptual Feature Integration (Lecture 5) Flashcards
Feature integration (or binding): what’s the problem?
Different visual features are processed in distributed manner across visual cortex (Tootell, Dale, Serena & Malach, 98).
- How does the cognitive system know which features go together?
Early studies of perceptual feature integration made use of the visual search task- Where’s Wally? is a relatively sophisticated visual search task.
VST- in the lab
Treisman & Gelade, 1980: aim = time taken to detect target.
- variables= what target looks like; number of distractors.
Feature integration theory
Claims that visual perception can be characterised by two functionally independent and sequential stages.
First stage is pre-attentive (can’t record/remember what you’ve attended- pre-conscious)- features are coded independently and in parallel using feature maps for colour, size, shape.
Second stage, cross-dimension processing is attentive- features from distinct feature maps are combined with each other via a window of attention that operates on a master map of locations (ie. when a window of attention falls on the shared spatial locations of the features).
Conjunction (attentive) search stage
- in order to combine the individuals features of an object, attention is required and selection occurs within a “master map” of locations.
- master map contains all locations which features have been detected, with each location in the ‘map’ having access to the multiple feature maps.
- when attention focused on particular point in map, features currently in that position are attended to and are stored in “object files”.
- if object is familiar, associations are made between the object and prior knowledge, which results in identification of that object.
FIT
Feature maps: for each type of visual feature (eg. depth, colour, orientation etc).
A master map of locations (MMLs): for mapping out the locations of stimuli.
An attentional spotlight: for combining info from feature maps via MMLs.
Feature (pre-attentive) search phase
Target differed by distracter by one feature (eg. colour or shape).
Task can be solved by checking feature maps only.
Conjunctive (attentive) search phase
Target differs from distractors by two or more features (ie. you can only find target based on correct combination of features).
Target can only be found by correctly integrating colour and shape features (two necessary features)- up one level in FIT.
Recruits spatial attention operating on the master map of locations.
FIT: principles summary
Feature pop-out arises in feature searches because distinctive feature value signalled on feature map. The maps are checked in parallel.
When distractors and target share feature values, the attentional focus must be narrowed so that a small number of elements can be checked serially.
Sequential selection is achieved via a window of attention that operated on the master map of locations.
Is spatial attention critical to perceptual feature integration?
Based on observations, some have suggested that spatial attention is the ‘glue’ that integrated features in visual perception.
Implications so far
FIT in combination with data from VST suggests that feature integration requires a window of attention operating on a master map of locations- occurs when more then one feature must be combined.
- when feature integration requires serially applied spatial attention.
Spatial attention: critical to feature integration
Visual search provides good evidence for the role of spatial attention in perceptual feature binding.
- Illusory conjunctions- predictions: without focal attention, binding errors (illusory conjunctions) should be observed.
Critically, in illusory conjunction the report of features is correct (done by checking feature maps), but the binding of these features is incorrect (something has gone wrong during the attentive stage).
Behavioural evidence: The flanker task
The occurrence of illusory conjunctions is high in situations where spatial attention is diverted and when stimuli are presented very briefly (so attention is unable to focus).
Treisman & Schmidt 82: four shapes in varying size, colour, and format arranged at corners of a square- flanked by two digits.
1) to report digits, then report features of one shape.
2) report features of one of the shapes (focused attention condition).
Found: illusory conjunctions commonplace (occur above 50% guessing rate), particularly in condition where flankers have to be reported- focused attention was prevented since the array was presented briefly and attention was spread across stimuli.
Flanker task altered to allow focused attention
Task: same stimuli, relevant object location cues 150ms before the rest of the array appeared.
Found: when attention could be focused, illusory conjunctions disappeared.
FIT: evaluation (pre-attentive search for conjunctions?)
Clearly FIT can explain large body of empirical evidence.
But- can’t explain:
- search for more complex objects (beyond simple primitive features- eg 3D) can be pre-attentive (which FIT would not predict).
- Enns & Resink, 91- presented search displays with objects that varied in 3D orientation- search times did not increase with number of distractors (search was solved pre attentively).
Treisman 93 argued back: conjunction ‘pop-out’ occurs when attentional window is broadly focussed, rather than narrowly focussed.
- global spread of attention can result in pop-out of some features.
FIT evaluation- influence of the nature of distractors
Duncan & Humphreys 89: suggest it’s the similarity between distractors is important in search too.
- they got ppts to find upright L amounts rotated Ts- this requires integration of elementary features (lines) to find the arrangement of lined to form a T or L.
FIT states this search should be attentive, but not all conditions were affected by set size.
When all distractors were homogenous (same orientation) search seemed to be pre-attentive (no influence adding more distractors)- FIT would not predict this.
When not homogenous, RTs increased with umber of distractors- suggesting visual search based on rapid rejection of the distractor group.
- this rejection process is harder when distractors are not homogenous, because they are harder to group- FIT cannot explain this finding.
