the feature-integration of attention_jing

A Feature-Integration Theory of Attntion

Anne Treisman & Garry Gelade (1980)

Jing Chen

Outline

• The Feature-Integration Theory of Attention

• Paradigms/Experiments• Visual search (Exps 1, 2, &3)

• Illusory conjunctions (Exp 4)

• Texture segregation (Exp 5, 6, & 7)

• Identity and location (Exp 8 & 9)

• Conclusion

The Feature-Integration Theory of Attention

• Features are registered early, automatically, and in parallel across the visual field, while objects are identified separately and only at a later stage, which requires focused attention to “glue” features together.

• Attention is necessary for the correct perception of conjunctions, although unattended features are also conjoined prior to conscious perception. • “Illusory conjunctions”

Dimension vs. Feature

• “Dimension” refers to the complete range of variation

• “Feature” refers to a particular value on a dimension

• Thus color and orientation are dimensions; red and vertical are features on those dimensions.

Experiment 1: Visual Search

• Purpose: • To compare search for disjunction targets and

conjunction targets;

• To explore the effect of extended practice on serial and parallel search;

• Among the distractors Tbrown and Xgreen

• Disjunction targets: a blue letter or an S;

• Conjunction targets: Tgreen;

• Setsize: 1, 5, 15, and 30

Disjunction Target

Find the blue letter

• Easy:

X T X T

X T T X

T X X X

T T X T

• Just as Easy: X T X T T T X T X T X X T X T T T X T T X X T X X X T X T X T X T X T T X T X T

Conjunction Target

Find the green “T”

• Hard:

X T X T

X T T X

T X X X

T T X T

• Even Harder: X T X T T T X T X T X X T X T T T X X T X T T X X X T X T X T X T X T T X T X T

Experiment 1: Results

• Conjunction: search time increased linearly

with setsize; search is serial and self-

terminating;• Disjunction: for positive displays, search times

were hardly affected by setsize; for negative

displays, the relationship is linear;


• The effect of extended practice:

• There is little indication of any

change in the pattern of results;

• no sign of a switch from serial to

parallel search.

Experiment 1: Discussion

• Focal attention, scanning successive locations serially, is the means by which the correct integration of features is ensured.

• When this integration is not required by the task, parallel detection of features should be possible.


• Purpose: to explore the relation between the discriminability of the features of a conjunction and the speed (slop) of detecting that conjunction as a target.• Compare • Difficult condition: a conjunction target in distractors

similar to it (T in X and T);

• Easy condition: distractors differed maximally from the target (O in O and N).


• The slopes in the difficult

discrimination are nearly three times

larger than those in the easy

discrimination

• but the linearity and the 1/2 slope

ratio is preserved across these

large differences.


• The search rates vary dramatically for easy and difficult conditions.

• In both easy and difficult conditions for the conjunction targets, the search were serial, self-terminating.

• As a result, we cannot say that search becaomes serial only when it is difficult.


• Purpose: to explore an alternative explanation for the difference between conjunction and disjunction targets.• attributes the difficulty of the conjunction condition to

the centrality of the target in the set of distracters.

• Replicate this aspect of the similarity by using unidimensional stimuli, with no need for checking conjunctions.


“Central”• The pattern of results is quite

different from that obtained

with the color-shape

conjunctions and disjunctive

features.

• When the intermediate target

is present, its detection doesn’t

depend on a serial check of

the distractor, which the

detection of the conjunction

did.

Experiment 4: Letter Search

• Purpose: to discover whether integrative attention is required even with highly familiar stimuli, e.g., letters

• Confusability of letters• Letters would be difficult to search when they are

similar in a wholistic way.• R/PB

• Sets of letters would be confusable if their features were interchangeable and could potentially lead to illusory conjunctions. • R/PQ

Experiment 4: Letter Search

• Purpose: to discover whether integrative attention is required even with highly familiar stimuli, e.g., letters

• The conjunction condition (with interchangeable features): R/PQ and T/IZ

• The similarity condition (with greater target/distractor similarity): R/PB and T/IY


• The ratio of positive to negative

slopes differed for the

conjunction and the similarity

conditions:

• For the conjunctions, it was

0.45, which is close to half

and suggests a serial self-

terminating search.

• For the similarity condition,

it was much lower (0.26).


• Letter search would be serial and self-terminating if the particular sets of distractor and target letters were composed of perceptually separable features which could be wrongly recombined to yield conjunction errors.

• Otherwise search could be parallel (although not necessarily with unlimited capacity and no interference).

Experiment 5: Texture Segregation

• Purpose: to investigates the “preattentive” segregation of groups and textures, which could guide the subsequent direction of attention.

• Five rows * five columns; card sorting task:• The color condition: OV|OV

• The shape condition: OO|VV

• The conjunction condition: OV|OV

OOVOOVOVVVVOVOVOOVVOOVVOV

The task was to sort the packs of cards into two piles, one containing cards with a horizontal and one with a vertical boundary.


• The difference between the two feature packs and the conjunction pack was qualitative and obvious.

• The color condition: OV|OV

• The shape condition: OO|VV

• The conjunction condition: OV|OV

•

• Suggesting that the boundary cannot be directly perceived in the conjunction condition and has to be inferred from attentive scanning.

Face-up Face-down

15.9 25.1

16.2 25.6

24.4 35.2


• Purpose: to discover whether the advantage of the feature boundary was due to• In the feature pack, only one dimension was relevant

• But the conjunction pack, require attention to both dimensions.

• Change the feature display into multiple-dimensional one: OΠ|OV (26.9 vs. 32.9 sec)

• Results:• The disjunctive features appear slightly less effective than single

features.

• The relevance of two dimensions rather than a single dimension can explain only a small fraction of the difference between features and conjunctions.


• Purpose: to see whether the distinction between features and conjunctions is equally crucial when the features are local components of more complex shapes rather than values on different dimensions (as in 5&6).

• The single feature conditions (short diagonal line): • PO/RQ (779 ms)

• EO/FQ (799 ms)

• The conjunction conditions:• PQ/RO (978 ms)

• FK/EX (1114 ms)

Again, what matters is

whether the boundary is

defined by a single feature or

a conjunction of feature

Experiments 8&9: Spatial Location

• Purpose: to test whether precise information about spatial location is available at the feature level, • by looking at the dependency between reports of

identity and reports of location on each trial.

• Difference between Exps 8 & 9:• In Exp 8, the presentation times of the arrays were

chosen to make sure the accuracy in each condition was 80%.

• In Exp 9, equal presentation times were used for features and conjunctions.

Experiments 8&9: Spatial Location

• All distractors were OX

• The targets:

• The disjunctive feature condition: H, H, X, or O

• The conjunction condition: X or O

• Dependent variable:• accuracy with brief exposures

OXOXOOOXOXHX

Experiments 8&9: Results

The conditional probabilities follow a very similar pattern.It seems likely that in order to focus attention on an item, we must spatially localize it and direct attention to its location.Feature localization is a special kind of conjunction task (feature and spatial location).

Conclusions

• All the data taken together support the feature-integration theory of attention.

• Separable features can be detected and identified in a early, parallel process;

• this process mediates texture segregation;

• locating any individual feature requires an additional operation;

• conjunctions of features require focal attention to be directed serially.

the feature-integration of attention_jing

Technology