BrainFinch
Articles

Visual search: why some targets pop out and others hide

A target pops out when it differs from everything around it by one simple feature, such as being the only red item among green ones. It hides when only a combination of features marks it, such as a red circle among red squares and green circles. Anne Treisman and Garry Gelade set out this distinction in 1980.

Target Hunt, the game above, moves from the first kind of search to the second as a round goes on.

Visual search is looking for one thing among others. The thing you want is the target and the rest are distractors.

In the laboratory a trial shows a display of items, and the person reports as fast as possible if the target is there or where it is. The number of items, called the set size, changes from trial to trial. The measure is how much each extra item adds to the response time. Plotted as a line, that is the slope, and Wolfe and Horowitz (2017) describe the slope as the measure of how efficient a search is.

A search is needed at all, they point out, because people cannot recognize everything in view at once.

What did Treisman and Gelade propose in 1980?

They proposed that seeing an object happens in two stages, and that only the second needs attention. The paper is “A feature-integration theory of attention”, published in Cognitive Psychology.

In the first stage, simple features such as color and shape are picked up automatically across the whole scene, before attention is directed anywhere. In the second, attention is directed to one location and the features found there are joined into a single object.

The theory makes a prediction about search. A target that has a feature nothing else has should be found from the first stage alone. Treisman predicted that such targets would pop out of a display. A target that can only be told apart by how its features combine needs the second stage, so attention has to be brought to items in turn.

A feature search has a target with one unique feature. A conjunction search has a target whose features are each shared with some of the distractors, so only the combination is unique.

Feature search Conjunction search
Example target A red circle A red circle
Example distractors Green circles Red squares and green circles
What marks the target Color alone Color and shape together
Effect of adding items Little or none Response time rises

The two columns have the same target. What changes is the crowd.

For feature search the textbook result is a flat line: response time stays about the same as items are added. Wolfe and Horowitz call a slope near zero the mark of pop-out. For conjunction search the line rises. The companion site to the Wolfe textbook Sensation & Perception gives typical values of about 10 to 30 milliseconds per added item.

Later work softened the sharp two-way split. Wolfe and Horowitz write that features can guide attention, and that it is possible to guide attention to more than one feature at a time. A searcher looking for a red circle can give priority to red items. They also cite Duncan and Humphreys (1989) for two rules: a target stands out more the more it differs from the distractors, and the more alike the distractors are to one another. Along basic feature dimensions, they add, search efficiency depends on the number of features shared by targets and distractors.

That last point is the one Target Hunt is built on.

What does a board in Target Hunt look like?

A chip at the top shows the target, and a square board below holds a scatter of tokens. Exactly one token is the target. You tap it for a point and a new board appears. A round lasts 30 seconds.

Each token has two features that we read from the game’s data file:

  • A shape, one of six: circle, square, triangle, diamond, star or cross.
  • A tone, one of two: solid in the game’s accent color, or a hollow outline in ink.

Six shapes and two tones make 12 possible kinds of token. The hollow tokens are drawn as outlines so that the two tones never differ by color alone.

Tokens sit in the cells of a grid, one per cell, and each is nudged off the center of its cell by a random amount so the board looks scattered. Some cells are empty.

A wrong tap leaves a cross on that cell for 450 milliseconds, and nothing can be tapped until it clears. The same board stays up until you find the target. The target changes with every new board, so the chip has to be checked each time.

How does Target Hunt build its crowd at each step?

The crowd is built from a step number: your level plus one for every two targets you have found in the round. Your level starts at 0, goes up after two rounds in a row with at least 80 percent accuracy, and comes down after a round under 50 percent.

The step sets three things: how many tokens are on the board, how big the grid is, and which features the distractors share with the target. The table is read from the generator code, and we ran the generator at each step to count the tokens.

Step Tokens Grid Distractors with the target’s shape Distractors with the target’s tone Other shapes in the crowd
0 5 3 by 3 0 4 1
1 7 3 by 3 0 6 1
2 9 4 by 4 0 8 2
3 11 4 by 4 0 10 2
4 13 4 by 4 0 12 2
5 15 5 by 5 7 7 1
6 17 5 by 5 8 8 1
7 19 5 by 5 9 9 1
8 21 5 by 5 10 10 1
9 and up 22 5 by 5 10 11 1

The code names three modes, and they line up with the rows.

At steps 0 and 1, every distractor is the same single shape, different from the target, in the target’s tone. Tone is the same across the whole board, so shape alone marks the target. This is a feature search with a uniform crowd, the easiest case under the Duncan and Humphreys rules.

At steps 2 to 4, the distractors are two different shapes, still all in the target’s tone. It remains a search by shape alone, but the crowd is no longer uniform and there are more tokens.

From step 5 the board is a conjunction search. About half the distractors have a different shape in the target’s tone. The rest have the target’s shape in the other tone. If the target is a solid star, the crowd is solid squares, say, and hollow stars. Neither “solid” nor “star” is enough. Only the two together pick out one token.

The token count goes up by two per step and stops at 22 on a 25 cell grid, which leaves three cells empty.

How do you find the target faster?

These tips come from how the boards are built.

  1. Check the chip for both features. From step 5 you need the shape and the tone, and a wrong tap costs 450 milliseconds of locked board.
  2. On conjunction boards, search one half. Pick a feature and look only at tokens that have it. At step 5 there are 15 tokens, but only 8 are in the target’s tone: the target and 7 others. Among those 8, the target is the only one with its shape. At step 9 the same move cuts 22 tokens to 12.
  3. You can split by shape instead. At step 5 there are 8 tokens with the target’s shape, and the target is the only one in its tone. Use whichever feature you find easier to pick out. This is the guidance Wolfe and Horowitz describe, applied by hand.
  4. On early boards, ignore tone. Up to step 4 every token on the board has the same tone, so it carries no information.
  5. Know when the change comes. A player at level 0 reaches step 5 after ten targets. A player at level 5 or above starts every round on conjunction boards.
  6. Sweep in order when a glance fails. The board is at most five rows of five, and tokens are offset inside their cells, so a row by row pass covers every cell once.

A Target Hunt score is the number of targets found in 30 seconds. It is a game result and does not measure anything about a person’s eyesight, attention or ability. Practice makes people faster at this game. Whether practice of this kind helps with anything else is not established, as we explain in what the research shows about brain training games.

For other kinds of search, Number Hunt has you find numbers in order on a shuffled grid, and we cover its background in Schulte table: what it is and how to play one online. Distractor Tap shows shapes one after another and asks for a tap on the target only, and Count Only has you count the matching shapes as they drift past.

Common questions

What is visual search?

Visual search is the act of looking for a target among other items, called distractors. In the laboratory it is studied by timing how long people take to find a target as the number of items on screen goes up.

What is the difference between feature search and conjunction search?

In a feature search the target differs from every distractor by one simple feature, such as color, and it tends to be found quickly however many items there are. In a conjunction search the target is defined by a combination of features that each appear on some distractors, and it takes longer as items are added.

What is feature integration theory?

It is a theory published by Anne Treisman and Garry Gelade in 1980. It says simple features such as color and shape are registered early and automatically, and that attention is needed to join the features at one location into a single object.

Sources

  1. Treisman, A. M., & Gelade, G. (1980). A feature-integration theory of attention. Cognitive Psychology, 12(1), 97-136.
  2. Wolfe, J. M., & Horowitz, T. S. (2017). Five factors that guide attention in visual search. Nature Human Behaviour, 1(3), 0058.
  3. Wikipedia: Feature integration theory
  4. Oxford University Press, Sensation & Perception companion site: Feature and conjunction search activity

Games in this article

Take all 62 games with you

The BrainFinch app plays offline, keeps your streak and reminds you when the daily challenge is ready.

Get it on Google Play