Color Is Processed Less Efficiently Than Orientation in Change Detection but More Efficiently in Visual Search

Predictions From the Boolean-Map Theory

The Boolean-map theory of visual attention proposes that momentary conscious access can be characterized as a Boolean map (Huang, 2010; Huang & Pashler, 2007; Huang, Treisman, & Pashler, 2007). Briefly, a Boolean map can contain a set of locations in a map but only one single “feature label.” Given this one-label limit, one can predict that when the stimuli consist of parts that are associated with different features, an observer can perceive either one part with a feature label or the spatial structure of all parts but with no feature label (see Figs. 1a and 1b). The information carried by the Boolean map includes both the feature label and the spatial structure of the map itself.

OPEN IN VIEWER

Fig. 1.

Illustration of the data format of Boolean maps in the context of the (a) color and (b) orientation stimuli used in the present study. A Boolean map can contain a set of locations in a map but only one single “label” per dimension to represent the features of the map. Given this one-label limit, one can predict that when the stimuli consist of parts that are associated with different features, an observer can perceive either one part with a feature label or the spatial structure of all parts but with no feature label. The map structure is therefore never useful when an observer is trying to perceive the colors of a few disks. However, when an observer is trying to perceive the orientations of a few bars, the map structure is especially useful in a change-detection task and moderately useful in a perceptual-discrimination task, but rarely useful in a visual search task.

For representing a feature such as the colors of disks (Fig. 1a), the map structure is obviously never useful. However, for representing the orientations of bars (Fig. 1b), the usefulness of the map structure depends on the task that needs to be accomplished. For perceiving a single bar (perceptual discrimination), both the map structure and the feature label can be used to complete the task, so the map structure is moderately useful. For memorizing a set of bars (change detection), the map can be readily used to hold the overall structure, so the map structure will be especially useful. For finding an orientation-defined target (visual search), the map structure, which contains no immediate information about individual features, is of very little use.

Following the above theory, if appropriate stimuli are chosen to ensure that disk colors and bar orientations are processed equally efficiently in a perceptual-discrimination task, then bar orientations will be processed more efficiently than disk colors in a change-detection task, but less efficiently in a visual search task (see Fig. 2). In the present study, I ran all three of these tasks to test this hypothesis.

OPEN IN VIEWER

Fig. 2.

Predictions about the relative efficiency with which colors and orientations should be processed across the three tasks. The information carried by the Boolean map includes both the feature label and the spatial structure of the map itself. The feature labels always provide information regarding both colors and orientations (yellow boxes). The map structure provides no useful information regarding colors. However, it provides especially useful, moderately useful, and not very useful information regarding orientations in, respectively, change-detection, perceptual-discrimination, and visual search tasks. These differences are indicated by the different sizes of the blue regions. Thus, if one chooses appropriate stimuli to ensure that disk colors and bar orientations are processed equally efficiently in perceptual-discrimination tasks, then orientations should be processed more efficiently than colors in change-detection tasks, but less efficiently in visual search tasks.

Method

Apparatus and stimuli

In all three tasks, stimuli were presented on a CRT monitor with a resolution of 1,024 × 768 pixels, and participants viewed the display from a distance of about 60 cm. Examples of the stimuli are shown in Figure 3a. I used two types of stimuli: one in which the salient feature was disk color and one in which it was bar orientation. Each stimulus type included four feature values. The color disks (diameter = 0.36 cm) could be red, green, blue, or yellow. The bars (length = 1.3 cm, width = 0.2 cm) could be oriented at 0°, 45°, 90°, or 135°.

OPEN IN VIEWER

Fig. 3.

Stimuli (a) and sample trial sequences (b) for each stimulus type across the three tasks. Stimuli consisted of disks, in which the salient feature was color, or bars, in which the salient feature was orientation. In the perceptual-discrimination task, a central cue was followed by a single stimulus, and participants had to report whether the salient feature of the stimulus matched that of the cue or did not. In the visual search task, participants saw two arrays on either side of a central cue and were asked to report on which side of the display a stimulus matching the salient feature of the cued item was located. In the change-detection task, participants were asked to report whether or not a memory and a probe display were the same.

As shown in Figure 3b, in two of the three tasks (visual search and change detection), the stimulus items were presented in eight locations and arranged as two 2 × 2 arrays on the left and right sides of the display; the center of each array was 2.91 cm away from the center of the display. In each array, the distance between items, both horizontally and vertically, was 1.46 cm. In the perceptual-discrimination task, the single target item was randomly presented in one of the eight locations of the two arrays.

Results

Performance in the three tasks is plotted in Figure 4. ¹ The y axis represents performance in each task measured on logarithmic scales. Performance in the perceptual-discrimination and visual search tasks is presented in terms of exposure-duration thresholds, and performance in the change-detection task is given in terms of estimates of working memory capacity (Cowan’s K, Cowan, 2001). The exposure-duration thresholds are presented inversely, because smaller thresholds imply better performance.

OPEN IN VIEWER

Fig. 4.

Mean performance as a function of task and stimulus feature. Performance for orientation stimuli is expressed relative to performance with color stimuli, which was set to 100%. Note that the y-axis is on a logarithmic scale. The numbers above the bars show the mean values of the dependent measure for each task (exposure-duration threshold for perceptual discrimination and visual search, and estimates of working memory capacity for change detection). Note that for exposure duration, a shorter threshold indicates better performance.

Performance in the perceptual-discrimination task was approximately equal for color and orientation. More interesting, performance was substantially better for orientation than for color in the change-detection task but dramatically worse in the visual search task.

With regard to the statistical indexes, the difference in effect size (Cohen’s d) for color and orientation was, respectively, −0.18, −1.95 (p < .0001), and 2.41 (p < .0001) in the perceptual-discrimination, change-detection, and visual search tasks. A pairwise comparison of the differences between the three tasks showed that Cohen’s ds for visual search versus change detection, visual search versus perceptual discrimination, and perceptual discrimination versus change detection were, respectively, 4.24 (p < .0001), 2.80 (p < .0001), and 1.08 (p < .02). These extremely large effects make it clear that disk colors and bar orientations are not processed in the same general way across different tasks. The pattern of effects indeed followed the predictions from Boolean-map theory.

Discussion

When perceptual discrimination is used as a baseline, the results can be summarized as follows: Bar orientations are (a) processed more efficiently than disk colors in change-detection tasks but (b) processed less efficiently in visual search tasks. This unique combination (i.e., opposite effects on visual search and change-detection tasks) provides strong support for the Boolean-map notion and renders most alternative accounts unlikely. For example, one may argue that color is an especially salient feature, and thus a color-defined target can be searched for more efficiently than a non-color-defined target. It would be difficult to explain using any such account why the efficiency with which color and orientation are processed is reversed in a change-detection task, especially given that visual attention and visual working memory are generally closely related (Awh & Jonides, 2001; Awh, Jonides, & Reuter-Lorenz, 1998; Downing, 2000; Olivers, Meijer, & Theeuwes, 2006; Pashler & Shiu, 1999).

The first finding seems to be intuitive: Change can be detected by monitoring the global spatial structure. Indeed, Alvarez and Cavanagh (2008) similarly found that working memory for bar orientation is better than it is for Gabor orientation. However, they attributed their results to “different codes for the boundary and surface features” (p. 346) rather than to the use of spatial structure. Therefore, the present study has filled an important theoretical gap.

The second finding, on the other hand, is more counterintuitive. Bar orientations are searched dramatically less efficiently than disk colors after they have been controlled in perceptual discrimination. The Boolean-map explanation for this finding is that the processing of even a single bar mainly relies on its spatial structure rather than its feature.

Beyond supporting the Boolean-map notion, the present study shows that the assumption about the generality of processing visual features is seriously flawed. For an earlier example, Kosslyn et al. (1989) reported that the speeds of categorical and coordinate judgments depend on the hemifield in which they are processed. The present study adds another large and sophisticated deviation to the records (see also Alvarez & Cavanagh, 2008). Complete knowledge of these deviations is important. For example, it can help researchers have better control over experiments. If one wanted to control for feature differences, one could arbitrarily use any of the three tasks in this study as a control and jump to very different conclusions. A complete knowledge of these deviations would help to avoid this problem.

The present study is clearly relevant to research on configural processing (e.g., Banks & Prinzmetal, 1976; Pomerantz, 2003; Pomerantz, Sager, & Stoever, 1977; Weisstein & Harris, 1974). However, there are also important differences between this study and previous ones. First, previous studies often focused on configurations of multiple elements, whereas the present study shows the importance of the spatial structure in even a single bar. Second, the critical configurations in these previous studies often led to special emergent features (e.g., 3-D cubes or holes). But in the present study, the stimuli rarely formed such emergent features. This perhaps explains why previous studies have usually found that the configuration helps when finding a target, whereas the present study found that, in some sense, the configuration hinders finding a target. The former reflects the effect of special configurations that form emergent features, whereas the latter seems to reflect the effect of those nonspecial configurations that do not form emergent features.