Attention Priority Map Explains Unusual Visual Search Phenomena

© iStock

Researchers from HSE University and Harvard have found that the grouping of multiple elements in a visual display does not affect the search speed for an element with a unique combination of features. The Guided Search theory predicted such results. The study is published in the Journal of Vision.

Visual search is quite a common task in everyday life. It isn’t difficult, for example, to find an object with a unique feature, such as a red letter (the target) among a number of blue letters (the distractors). This is called a feature search. It is more difficult to find an object that differs from others by a combination of features — for example, a red letter 'L' among red letters ‘O’ and blue letters ’T'. In such situations, you perform a conjunction search.

In the picture above, grouping is complicated by the variety of letter colors. In each case, you must find the red letter ‘L’

Conjunction Search Theories

Scientists have long been known that the speed of searching for a target with a single feature depends on how easy it is to group the objects in a display. When the letters are clearly divided into two color categories (upper left image), the red element immediately grabs attention (“pops out”) among the blue elements - because all blue elements are clearly grouped together and the red one stands apart. However, if the letters have various shades of color from red to blue (lower left image), it is more difficult to find the red letter, because some distractors are now ready to be grouped with the target and because some of the other distractors are intermediate between red and blue. 

Intuitively, it would seem that in search for a combination of features, a particular element could be found faster in clear grouping (upper right image). This is the result predicted by search theories with a sequential mechanism of selection, which posit that the search for a combination of features is a two-step process. When letters are grouped easily by color, you first select all the red letters and search only for differences in shape, making the task twice as easy. However, if the collar are not easily grouped into two colors, it would be more difficult to distinguish between groups and the search would take longer.

In contrast, a search theory with a simultaneous mechanism of selection—namely, the Guided Search theory developed in the late 1980-s by one of paper’s authors, Jeremy Wolfe—proposes just the opposite. It argues that information about all relevant features is used simultaneously when directing one’s attention towards a potential target. That is, if you are looking for a red letter, your visual system will simultaneously activate locations with both red letters and letters of a certain shape. Thus, your attention will focus first on those places where both the color and shape are as similar to the goal as possible.

Verifying Theories by Experiment

The scientists conducted a series of experiments to test the predictions of these two theories. In the first, subjects were asked to search for a line of a particular color and orientation among other lines — for example, the steep red line in the pictures below. The difficulty of categorizing lines by their features varied among the trials: in some trials, the lines were clearly grouped by color and orientation, while in others, the differences in color and orientation were less distinct, making it more difficult to divide them into clear groups. 

An example of stimuli in one of the experiments, where the ease of categorizing elements into groups varies based on their characteristics — orientation and line color

The sequential mechanism predict that subjects can find the target faster when the grouping of lines is easy (the leftmost picture) and that they will have increasing difficulty as the stimuli become more variable and thus harder to group (the rightmost picture). However, the subjects completed the task with an identical speed for all four types of images. This result is surprising because not only is it counterintuitive, but also earlier research demonstrated that grouping influences the speed of searches for a single feature.

The Attention Priority Map as a Main Search Controller

This is explained by the search theories with a simultaneous selection mechanism, and particularly Guided Search theory. The distribution of features over the display automatically produces a sort of maps of feature activation (e.g. color maps or orientation maps), in which the activation level of at each location indicates the similarity between a feature presented at that location and that of the target. The outputs of these two Feature maps are then summed up on the Attention Priority Map whose activation defines how likely each area contains all target features. In other words, attention would likely first go to wherever activation is highest. This mechanism makes it equally easy to pay attention to a target regardless of the complexity of grouping, because there is only one place having both target features at once.

The attention priority map explains why good or poor grouping has no effect on search speed. No matter how many areas of a feature map have similar activation levels, when they are pooled, the target location is singled out clearly on the attention priority map

To further test the proposed explanations, the researchers conducted another experiment where, in addition to search for a feature combination, subjects were asked to search for a line based on only one feature, its unique orientation or color. It turned out that in the poor grouping conditions, searching for a combination of features (for example, a white vertical line among lines of different colors and orientations) can be faster than searching for a single feature (for example, a white line among lines of different colors or a vertical line among lines of different orientations). 

Seemingly, this does not fit into the visual search theory at all, and even violates one of its ‘laws’ — namely, that the search for a single feature should be easier than a search for a combination of features. However, the attention priority map hypothesis of Guided Search provides an explanation for this unusual finding: when searching for a single feature, many parts of the map show similar levels of activation, whereas when searching for a combination of features, the location containing the target object has the highest level of activation.

The attention priority map shows why, given hard categorization, search for a combination of features can be easier than search for a single feature. In searches for a single feature, multiple places on the attention priority map have similar activation levels, but in searches for a combination of features, a single area on the map shows a higher level of activation than all the others

Vladislav Khvostov, one of the study’s authors

Everyone does a lot of visual searches every day. It is very rare that the object we are looking for has one unique feature: more often, it has a certain combination of several features. In this study, we have demonstrated that the search for a combination of features works just as efficiently, regardless of how good the grouping of the elements is. This is made possible by building an attention priority map that collects the information about several features simultaneously.