Color onsets and offsets, and luminance changes can cause change blindness

Transcription

1 Perception, 2006, volume 35, pages 1665 ^ 1678 DOI: /p5599 Color onsets and offsets, and luminance changes can cause change blindness James G Arrington, Daniel T Levinô, D Alexander Varakin Department of Psychology and Human Development, Vanderbilt University, Peabody College, Box 512, 230 Appleton Place, Nashville, TN , USA; daniel.t.levin@vanderbilt.edu Received 16 February 2005, in revised form 29 December 2005 Abstract. It has recently been demonstrated that people often fail to detect between-view changes in their visual environment. This phenomenon, called `change blindness' (CB), occurs whenever the perceptual transient that usually accompanies a change is somehow blocked, or made less salient. In the well-known flicker paradigm, the transient is blocked by inserting a blank screen between the original and changed scenes. We tested whether transients that do not involve the appearance or disappearance of visual objects would also produce CB. Therefore we tested whether the appearance or disappearance of color information, and increments or decrements in luminance, could cause CB. In three experiments, subjects searched for changes in natural scenes. We found that both color transients and luminance transients significantly reduced change detection (by 30%) relative to a no-transient condition. 1 Introduction Recently, researchers have been exploring a wide variety of failures of visual awareness, ranging from failures to detect visual changes or failures to detect unexpected stimuli, even when they are bizarre or noxious (eg Simons and Chabris 1999; Wayand et al 2005). One of these failures is referred to as `change blindness' (CB), and it occurs when people are unaware of visual changes that occur across views (for reviews see Simons and Levin 1997, 2003; Rensink 2002). For example, CB can occur when subjects view a short film in which visual properties, such as the actors' clothing, change between shots. Even though the changes appear salient once they are pointed out, almost all subjects miss them, especially when they are not looking for changes (Levin and Simons 1997). The key to producing CB is to somehow block the perceptual transient that would normally call attention to a change. Often this is done by adding another broader transient such as a flicker (Rensink et al 1997), or a set of object onsets (O'Regan et al 1999), simultaneous with the change. Thus, the change-detection task can inform research that explores exogenous capture of attention in a situation where two transients compete. Typically, attentional capture is revealed as a transient-imposed cost, or benefit to a primary task such as visual search or probe detection. One characteristic of these tasks is that they typically focus the subject's attention on a very specific well-defined target dimension or category, and this focus may cause similar restrictions on the transient that can capture attention (Simons 2000). In contrast, a change-detection task, particularly one which involves natural scenes, requires subjects to have a very broad focus, because any of a large number of properties or objects may change. This makes the changedetection task particularly suitable for studying the breadth of transients that may capture attention. In these experiments, we first explored the degree to which the sudden addition or removal of color, and changes in overall luminance, could induce CB. 2 How can one block transients to produce CB? CB has been produced in a wide variety of ways. Initial research demonstrating CB involved changes that were synchronized with saccades. Thus, subjects' eye movements ô Author to whom all correspondence should be addressed.

2 1666 J G Arrington, D T Levin, D A Varakin were monitored while they viewed text or natural scenes, and visual changes were introduced after the initiation of a saccade (McConkie and Zola 1979; Grimes 1996; Hollingworth and Henderson 2000). This technique is highly effective at masking the changes because the retinal `smear' that accompanies a saccade effectively competes with the change-induced transient. However, the technique is quite demanding technically, so researchers interested in exploring CB have designed a variety of ways to mask the change without the need for eye tracking. The first, and most widely used, of these methods has been to insert sudden visual onsets and/or offsets at the same time as the change. In the `flicker paradigm' (Rensink et al 1997), subjects view two different versions of a scene. For example, one version of the scene might have a building in the background which is edited out of the other version. The versions are switched back and forth, separated by a blank grey screen. The fact that the grey screen causes the scene to disappear and reappear allows the transient caused by the change to be lost in the more global transient that occurs in the disappearance and reappearance. Moreover, these onsets and offsets do not need to entail the appearance and disappearance of the entire image. O'Regan et al (1999) found that the sudden addition of a set of spots to the screen at the time of the change is sufficient to cause CB, even though none of the spots obscures the change itself. Another means of creating a sufficient transient is to simply move the image diagonally simultaneous with the change (Blackmore et al 1995), or, as in motion-picture edits, to make the change simultaneous to the change in viewpoint (Levin and Simons 1997). Even simpler, change detection can be disrupted by briefly occluding the changing objects during the change (Simons and Levin 1998). Finally, changes can be made gradually by the use of `fade-ins' that occur over the course of several seconds (Simons et al 2000). Recently, Turatto et al (2003) reported that CB could be induced by sudden contrast inversions that do not include flickers. In their experiments, subjects viewed black-andwhite images of natural scenes that were alternately presented in their normal and contrast-reversed forms. Contrast inversions are different from previous methods of inducing CB in that they involve no onsets or offsets of objects (which one could argue are necessary to attract attention away from the change; see Jonides and Yantis 1988) and do not require that the scene be off-screen when the change is made. However, contrast inversion still represents a substantial and sudden transformation in the apparent shape of objects. For example, faces are more difficult to recognize when contrast-inverted, which suggests that the inversion interferes with the subject's ability to extract form or configural information from the face (Kemp et al 1996). Although it is plausible that Turatto et al's inversion disruption was successful because it involved form-based or shape-based transients, there are several reasons to believe that this is not the case. For example, it is well known that quick changes in luminance strongly activate both ventral/parvocellular and dorsal/magnocellular parts of the visual system (Livingstone and Hubel 1988), and a number of studies have demonstrated that these transients can powerfully capture attention (see Simons 2000 for a review). However, the attentional-capture capability of other non-form transients such as sudden changes in color is less clear. Some studies have reported that these changes do attract attention (Gellatly et al 1999; Snowden 2002; Lu and Zhou 2005), whereas other studies have provided evidence that they do not. For example, Cole et al (2005) argued that Snowden (2002) had observed color-based attentional capture because the color transient in his study was confounded with an object onset. When Cole et al removed the object onset, color changes did not capture attention. One limitation of many of these findings is that they often start from the assumption that a given cue either does or does not attract attention across the board. In contrast, other researchers have argued that attentional capture instead reflects the match between the nature of the target-detection task and the nature of the irrelevant stimulus.

3 Color and luminance changes 1667 For example, Folk et al (1992) found that when a color singleton was the irrelevant stimulus in an onset-detection task it did not attract attention, but when the same singleton was presented in the context of a color-target detection task, it did attract attention. On the basis of this and other findings, they argued that attentional capture occurs when task parameters focus subjects' attention on the specific feature dimension that changes. This is particularly important in settings where the task requires searching for a change in a natural scene. Given that the task of detecting these changes may require subjects to search for changes in any one of a range of features, such as form, luminance, or color, it is possible that the transients capable of interfering with change detection will be similarly broad. The present experiments were designed to test the hypothesis that color onsets and offsets are sufficient to reduce change detection below a no-transient baseline. In experiment 1, we compared a color-onset condition with a no-flicker condition, and with a no-transient condition. In experiment 2, we added a color-offset condition, along with a no-transient grey-scale condition to test the degree to which gray-scale images in themselves reduce change detection. In experiment 3, we tested luminance changes to compare their ability to reduce change detection. 3 Experiment 1 Experiment 1 was designed to determine whether the onset of color alone could mask a change. We therefore compared change detection under four conditions: (i) the change was masked both by a color onset (the pre-change scene was presented in grey scale, and the post-change scene was presented in color) and by a 100 ms flicker between the pre-change and post-change scenes; (ii) the change was masked by the onset of color, with no flicker; (iii) the change was masked by traditional means, with both pre-change and post-change scenes presented in color, and separated by a flicker; and (iv) the change was not masked by flicker or by color onset. We therefore sought to determine the degree to which the color onset would cause CB relative to a condition that produced no transient and to one that contained a traditional large-scale disruption. 3.1 Method Participants. Twenty-nine undergraduate students from Kent State University participated in this experiment. All were enrolled in General Psychology, and none of them had previous experience with other CB experiments. Two of the participants were excluded because they did not follow instructions when making their responses (they indicated that the color transients themselves were changes). Three of the participants were excluded because they did not have normal or corrected-to-normal vision based on self-report. The analyses were conducted with the remaining twenty-four participants Apparatus and stimuli. Stimuli for this experiment were created from thirty-two color pictures of natural scenes, which were collected from The Big Box of Art (Hemera Technologies, Hull, Quebec). Pictures were selected based on the presence of an object, objects, or part of an object, that could be easily altered so that the changed versions would look as natural as the original versions. Three versions of each scene were created with Adobe PhotoShop 4.0 (Adobe Systems, San Jose, CA): pre-change scenes presented in 16-bit color, post-change scenes presented in 16-bit color, and prechange scenes presented in 8-bit grey-scale. We created a variety of changes but did not include any color changes. Of the 32 changes, 9 were additions or deletions of parts, 7 were changes to the height or width of objects, 5 were additions or deletions of whole objects, 2 were exemplar substitutions (eg replacement of an object with another object from the same basic-level category), 6 were changes to the surface pattern of an object, and 3 were rotations of an object (see figure 1). The conversion from color to grey-scale was done with the `change mode' command in Photoshop.

4 1668 J G Arrington, D T Levin, D A Varakin Addition/delection of object parts (9 scenes) Changes in height/width of objects or their parts (7 scenes) Addition/deletion of objects (5 scenes) Exemplar substitutions (2 scenes) Changes to surface pattern (6 scenes) Figure 1. Sample changes. Rotation changes (3 scenes)

5 Color and luminance changes 1669 This command achieves the change by computing a weighted mean of R, G, and B bit values. The weights reflect the Rec. 709 standard based on the typical computer monitor (Y 709 ˆ 0:2125R 0:7154G 0:721B; see Poynton 1998). All pictures were presented on Mac OS computers with 15-inch Sony fx100 CRT monitors set at a resolution of pixels (75 Hz). Luminance of the color and grey-scale versions of each scene was also measured directly from one of the monitors with a Minolta LS110 luminance meter. Each scene was sampled in four locations, one in each quadrant, and the average difference between the color and black-and-white scenes across all scenes was 0.20 cd m 2 (SD ˆ 0:55 cd m 2 ), and the average absolute difference between the versions was cd m 2 (SD ˆ 0:421 cd m 2 ). Each scene measured cm cm, and was presented at the center of the 21 cm628 cm screen at an unconstrained viewing distance of 60 cm. Thus, each scene subtended approximately deg (horizontal) deg (vertical) of visual angle Design and procedure. The two factors of interestöcolor onset versus no color onset and flicker versus no flickeröwere crossed, yielding four experimental conditions. The four conditions were: (i) color onset/flicker: grey-scale pre-change scene to a color post-change scene, separated by a flicker; (ii) color onset only: grey-scale prechange scene to a color post-change scene, without a flicker; (iii) flicker only: color pre-change scene to a color post-change scene, separated by a flicker; (iv) no transient: color pre-change scene to a color post-change scene, without a flicker. For conditions (i) and (iii), the pre-change scene was presented for 2000 ms, followed by a 100 ms `flicker' to mask the change, and then by the post-change scene. For conditions (ii) and (iv), the post-change scene immediately replaced the pre-change scene. For all four conditions, the post-change scene remained present until the participant made a response and was ready to move on to the next trial. The original thirty-two pictures were randomly divided into four sets of eight to serve in the four different conditions (the assignment of pictures to conditions was rotated across subjects so that all pictures served in all conditions). Participants sat approximately 61 cm from the computer screen, and were run in small groups ranging in size from 1 to 4. Participants were instructed that there would be a change on every trial, and that their job would be to detect what had changed in a variety of scenes. On each trial, they were instructed that they would see an initial version of the scene before the change, which would occur after a few seconds. They were then instructed that on some trials there would be a brief flicker before the change, and that on other trials there would be no flicker before the change. They were also instructed that either both scenes would be in color or the first scene would be in black-and-white and the second scene would be in color. Participants were explicitly informed that switching the image from black-and-white to color did not count as a change in the scene. After each trial, participants indicated whether or not they detected the change: first on the computer, and then on a response sheet. On the computer, participants were told to press the `Y' key if they detected the change or to press the `N' key if they did not. This response cue appeared simultaneously with the post-change scene. The response sheet was numbered 1 to 32 with a `Y' and `N' next to each number; each number also had a blank space labeled ``Pre'' and ``Post''. Participants were instructed to circle `Y' if they detected the change or to circle `N' if they did not. In addition, if they detected the change, the participants used the response sheet to write a description of the pre-change object and of the post-change object. This use of the keyboard response in addition to the response sheet was intended to allow for reaction-time measurements (which proved unilluminating, and are not reported here), to push subjects to make their change-detection responses without extensive deliberation, and to ensure accurate coordination between the computerized stimulus order files and the response sheets.

6 1670 J G Arrington, D T Levin, D A Varakin 3.2 Results Detected changes. Two independent raters scored the response sheets to verify that `Y' responses corresponded to correct change detections. The three criteria for scoring responses were: (i) the location of the changing object, more specifically that indicating the top/bottom or right/left side of the screen; (ii) the name of the changing object; and (iii) the correct type of change (appearance, disappearance, or movement). For each trial, change detection was scored as correct if the location and/or name of the changing object were correctly identified. If the type of change was correctly identified, but the location and/or name of the changing object were not. For the color-onset flicker condition, changes were detected on 14% (SD ˆ 14%) of the trials. For the color-onset-only condition, changes were detected on 35% (SD ˆ 14%) of the trials. For the flicker-only condition, changes were detected on 17% (SD ˆ 13%) of the trials. For the no-transient condition, changes were detected on 68% (SD ˆ 16%) of the trials. In a 2 (color onset versus no color onset)62 (flicker versus no flicker) withinsubjects ANOVA, there was a main effect both for color (F 123, ˆ 58:03, MSEˆ 1:3%, p 5 0:001) and for flicker (F 123, ˆ 137:35, MSE ˆ 2:2%, p 5 0:001). The interaction was also significant (F 123, ˆ 58:06, MSEˆ 0:9%, p 5 0:001; see figure 2). Pairwise comparisons confirmed that significantly fewer changes were detected in the color-onset-only condition than in the no-transient condition (t 23 ˆ 9:559, p 5 0:001), whereas significantly more changes were detected in the color-onset-only condition than in either of the flicker conditions (color onset only versus flicker only: t 23 ˆ 4:813, p 5 0:001; color onset only versus color onset/flicker: t 23 ˆ 6:244, p 5 0:001). The flicker conditions did not differ significantly from each other (t 23 ˆ 1:187, p 4 0:10). 100 Experiment 1 Experiment 2 Experiment 3 80 Change detection=% color onset/flicker color onset flicker no transient color onset color offset 3.3 Discussion In experiment 1, color onsets clearly increased the prevalence of CB, but not as much as a flicker. The color onsets reduced change detection by 33% relative to the notransient baseline, whereas the more traditional flickers reduced change detection by 51%. no transient color Condition Figure 2. Percentage change detection in experiments 1 ^ 3. no transient grey-scale luminance increase luminance decrease flicker no transient

7 Color and luminance changes 1671 One unusual aspect of the data is that change detection with no transient was only 68%. According to most accounts, these conditions should produce very high rates of change detection. One plausible explanation for the relatively low rates of change detection might be that at least some identification of the changing objects was required in order for a response to be counted as a hit. Therefore, we may be measuring change identification rather than change detection, and some authors have argued that the two are different (Fernandez-Duque and Thornton 2000, 2003). On the other hand, other researchers have observed similar patterns of results both for change identification and for change detection (Mondy and Coltheart 2000), and so it may be that change detection is simply the result of a more liberal coding criterion than change identification. In any case, the pattern of results for experiment 1 is similar even when all positive responses are counted as hits: color onset/flicker (M ˆ 33%; SD ˆ 24%); color onset only (M ˆ 52%, SD ˆ 18%); flicker only (M ˆ 28%, SD ˆ 20%); no transient (M ˆ 79%, SD ˆ 14%). The increased rate of change detection with this more liberal criterion could be caused by a number of factors. Most likely, subjects may sometimes have experienced a near-threshold change detection but were mistaken about its location. This possibility is reinforced by findings that changes can be detected by means of temporary, volatile representations that are partially disrupted by the post-change scene such that they remain, but lack strong information about specific changing properties or locations (Beck and Levin 2003). As mentioned above, the alternative is that the additional change detections are simply false alarms, and without no-change catch trials there is no good way of distinguishing these possibilities. One potential problem with experiment 1 is that the response cue may have acted as a distractor on each trial. On the computer, the response cue, ``Did you see anything change? Press the `Y' key if you did, or the `N' key if you did not'', appeared at the same time as the post-change scene and may have lowered change-detection rates overall. This issue will be eliminated in experiments 2 and 3. 4 Experiment 2 In experiment 2, we replicated and extended the results of experiment 1 by testing both color onsets and color offsets, and by adding an all-grey-scale no-transient condition. The color transient effect in experiment 1 was large, but it is possible that it occurred because subjects found it relatively difficult to search and/or parse the grey-scale images. Therefore, we added the all-grey-scale no-transient condition to provide a comparison with an all-color non-transient condition. If the grey-scale images were simply more difficult to process, then there should be a large reduction in change detection relative to color scenes when they are presented under no-transient conditions as well. In addition, we added a color-offset condition to determine whether onsets are special in attracting attention away from the change, or if offsets will serve the same purpose. 4.1 Method Participants. Thirty-seven undergraduate students from Kent State University participated in the experiment. All were enrolled in General Psychology, and none of them had previous experience with other CB experiments. Three participants were excluded because they did not follow instructions when making their responses. The analyses were conducted with the remaining thirty-four participants Apparatus and stimuli. Stimuli were similar to those used in experiment 1, with the exception that a set of 8-bit grey-scale post-change scenes was created to allow presentation of color offsets Design and procedure. Procedures were similar to those followed in experiment 1, with the exception that a color-offset condition and an all-grey-scale condition were

8 1672 J G Arrington, D T Levin, D A Varakin added, and the flicker conditions were eliminated. Experiment 2 was therefore a 262 cross of pre-change scene color (black-and-white versus color) and post-change scene color (black-and-white versus color). The four conditions were: (i) color onset: grey-scale pre-change scene to color post-change scene; (ii) color offset: color pre-change scene to grey-scale post-change scene; (iii) no transient color: color pre-change scene to color postchange scene; (iv) no transient grey-scale: grey-scale pre-change scene to grey-scale post-change scene. For all four conditions, the pre-change scene was presented for 2000 ms, followed immediately by the post-change scene. In experiment 2, participants were again asked the question ``Did you see anything change? Press the `Y' key if you did, or the `N' key if you did not''. However, the question appeared 2000 ms after the post-change scene appeared, instead of simultaneously as in experiment 1. For all four conditions, the post-change scene was present until the participant made a response and was ready to move on to the next trial. Participants were given a more detailed set of instructions for experiment 2. Participants were again instructed that there would be a change on every trial, that their job would be to detect changes in a variety of scenes, and that switching the image from black-and-white to color, or from color to black-and-white, did not count as a change in the scene. In addition, participants were given some examples of possible changes to a post-change scene, such as additions, deletions, size, and/or changes to the location of an object, objects, or parts of an object. These examples were provided to reinforce that changes in overall color did not count as a change. Participants were also instructed that there would not be any changes in color to the object, objects, or parts of an object in a post-change scene. They were told that there would be trials where: (i) both the pre-change and the post-change scenes would be in color; (ii) both the pre-change and post-change scenes would be in black-and-white; (iii) the pre-change scene would be in black-and-white and the post-change scene would be in color; and (iv) the pre-change scene would be in color and the post-change scene would be in black-and-white. Change detection was measured in the same way as in experiment Results Detected changes. Two independent raters scored the response sheets from this experiment by means of the same criteria as those employed in experiment 1. For the color-onset condition, changes were detected on 40% (SD ˆ 20%) of the trials. For the color-offset condition, changes were detected on 49% (SD ˆ 20%) of the trials. For the no-transient-color condition, changes were detected on 83% (SD ˆ 13%) of the trials. For the no-transient-grey-scale condition, changes were detected on 79% (SD ˆ 19%) of the trials. The pattern of results for experiment 2 is similar, even when all positive responses are counted as hits: color onset (M ˆ 51%, SD ˆ 23%); color offset (M ˆ 66%, SD ˆ 23%); no transient color (M ˆ 89%, SD ˆ 13%); no transient grey-scale (M ˆ 88%, SD ˆ 14%). A 262 within-subjects ANOVA was conducted with pre-change scene color (black-and-white versus color) and post-change scene color (black-and-white versus color) as factors (note that this ANOVA is possible because there were no flicker conditions, so the transients were embedded in a 262 cross of the presence of color in the pre-change scene and the post-change scene). The main effect for the color of the pre-change scene was significant (F 133, ˆ 4:452, MSEˆ 3%, p 5 0:05), but the main effect for the color of the post-change scene was not (F 133, ˆ 1:848, MSEˆ 1:6%, ns). The main effect for the pre-change scene color reflects better change detection when the pre-change scene was in color (66% detection) than when it was in grey-scale (60% detection). The interaction between pre-change scene color and post-change scene color was significant (F 133ˆ, 131:691, MSEˆ 3:4%, p 5 0:001; see figure 2). Pairwise comparisons revealed that both color onsets and offsets disrupted change detection relative to no-transient conditions. Change detection in the color-onset

9 Color and luminance changes 1673 condition was significantly worse than in the no-transient-grey-scale condition (t 33 ˆ 9:862, p 5 0:001), and change detection in the color-offset condition was worse than in the no-transient-color condition (t 33 ˆ 9:082, p 5 0:001). Change detection in the color-offset condition was better than in the color-onset condition (t 33 ˆ 9:082, p 5 0:012), and there was no difference between the no-transient conditions (t , ns). All comparisons between color-transient conditions and no-transient conditions in experiments 1 ^ 3 are significant with Bonferroni a posteriori tests. 4.3 Discussion Relative to both the no-transient-grey-scale baseline and to the no-transient-color baseline, both the onset and the offset conditions reduced change detection by at least 30%. Although there was an attenuation of change detection when the pre-change scene was in grey-scale, the effect was much smaller (6%). Therefore, it is clear that our color transients caused CB, and that the effect cannot be attributed solely to the difficulty of processing the grey-scale images. We should note that overall the base rate of change detection in this experiment does appear to be higher than that in experiment 1. This could have been caused by delaying the response cue, but we are hesitant to make between-experiment comparisons because the difference is relatively small, and not the main focus of this experiment. Thus, differences in subject populations recruited at different times during the semester could have easily caused this. In addition, we point out that the rate of change detection in the flicker-only condition of experiment 3 was just as low as that in experiment 1, despite the delayed response cue in that experiment. 5 Experiment 3 In experiments 1 and 2 we have argued that color onsets and offsets may be part of a relatively broad set of visual transients capable of causing CB. In experiment 3, we test whether luminance changes can also cause CB. Clearly, changes in luminance constitute a large part of the signal for many changes, and much research has explored visual channels dedicated to processing sudden changes in luminance (eg the magnocellular pathway; for a review see Shapley 1995). Furthermore, early research on attentional capture used changes in the luminance of an existing object as a cue (Posner et al 1980), and in certain circumstances sudden luminance changes may be even more powerful in attracting attention than color changes (Theeuwes 1995; but see Gellatly et al 1999). Therefore, luminance changes would also appear to be a good candidate to mask changes. In experiment 3, we directly compared the degree to which luminance changes and flickers cause CB in color images. 5.1 Method Participants. Twenty-four undergraduate students from Kent State University participated in this experiment. Sixteen of the students were enrolled in General Psychology and received extra course credit for their participation. The remaining eight students signed up through flyers and were paid US $5.00 for their participation. None of the twenty-four students had previous experience with other CB experiments. One participant was excluded because he/she did not follow the instructions when making his/her responses. The analyses were conducted with the remaining twenty-three participants Apparatus and stimuli. Stimuli were similar to those in experiment 1 and experiment 2, with the exception that sets of bright color pre-change and post-change scenes were created by increasing the luminance by 11.7%. This level of adjustment was selected to avoid significant loss of highlight detail. As in experiment 1, the luminance difference between the versions of each scene was also measured with a luminance meter. The mean difference was 9.80 cd m 2 (SD ˆ 3:30 cd m 2 ).

10 1674 J G Arrington, D T Levin, D A Varakin Design and procedure. The procedures were similar to those used in experiment 1 and experiment 2, with the exception that a luminance-increase condition and a luminancedecrease condition were added, both without a flicker. All scenes were presented in color. Therefore, the four conditions were: (i) luminance increase: color pre-change scene to color post-change scene with a luminance increase; (ii) luminance decrease: color pre-change scene with a luminance increase to color post-change scene; (iii) flicker only: color pre-change scene to color post-change scene, separated by a flicker; (iv) no transient: color pre-change scene to color post-change scene, without a flicker. For condition (iii), the pre-change scene was presented for 2000 ms, followed by a 100 ms `flicker' to mask the change, and then by the post-change scene. For conditions (i), (ii), and (iv), without a flicker between, the post-change scene immediately replaced the pre-change scene. As with experiment 2, participants were again asked the question ``Did you see anything change? Press the `Y' key if you did, or the `N' key if you did not''. However, the question appeared 2000 ms after the post-change scene appeared, instead of simultaneously as in experiment 1. For all four conditions, the post-change scene was present until the participant made a response and was ready to move on to the next trial. Participants were given a set of instructions for experiment 3 that were very similar to those in experiment 2. Participants were again given examples of the kinds of changes that would occur to a post-change scene, and were explicitly informed that switching the brightness of the image did not count as a change in the scene. They were instructed that there would be trials where: (i) the pre-change scene would be brighter than the post-change scene; (ii) the post-change scene would be brighter than the pre-change scene; and (iii) both the pre-change and the post-change scenes would be of equal brightness. Change detection was measured in the same way as in experiments 1 and Results Detected changes. Two independent raters scored the response sheets from this experiment by using the same criteria as those employed in experiment 1 and experiment 2. For the luminance-increase condition, changes were detected on 59% (SD ˆ 17%) of the trials. For the luminance-decrease condition, changes were detected on 51% (SD ˆ 16%) of the trials For the flicker-only condition, changes were detected on 14% (SD ˆ 11%) of the trials. For the no-transient condition, changes were detected on 85% (SD ˆ 13%) of the trials. The pattern of results for experiment 3 is similar, even when all positive responses are counted as hits: luminance increase (M ˆ 64%, SD ˆ 20%); luminance decrease (M ˆ 57%, SD ˆ 18%); flicker only (M ˆ 26%, SD ˆ 16%); no transient (M ˆ 86%, SD ˆ 13%). A within-subjects one-way ANOVA of the four conditions was significant (F 366, ˆ 108:64, MSEˆ 1:8, p 5 0:001; see figure 2). Pairwise comparisons revealed that change detection was worse in the flicker condition relative to luminance-increase (t 22 ˆ 9:898, p 5 0:001), and luminance-decrease conditions (t 22 ˆ 8:770, p 5 0:001). However, change detection was better in the no-transient condition relative to the luminance-increase (t 22 ˆ 8:321, p 5 0:001) and luminancedecrease conditions (t 22 ˆ 8:853, p 5 0:001). There was also a difference between the luminance-increase and luminance-decrease conditions (t 22 ˆ 2:091, p ˆ 0:048), reflecting better change detection in the luminance-increase condition. 5.3 Discussion Luminance increases and luminance decreases clearly increased the prevalence of CB, but not as much as a flicker. The luminance increases and luminance decreases reduced change detection by 26% and 34%, respectively, relative to the no-transient baseline, whereas the more traditional flickers reduced change detection by 71%.

11 Color and luminance changes 1675 Given that luminance transients clearly produced CB, we need to look back at the results from experiments 1 and 2 to ensure that the effects of color changes were not entirely due to luminance transients. Although we confirmed that the overall luminance was matched between the color and grey-scale versions of the scenes, this does not ensure that all of the regions in each scene were luminance matched. Indeed, it would be extraordinarily difficult to demonstrate that color and grey-scale versions of natural scenes were matched for luminance throughout. Therefore, instead of attempting to equalise the stimuli, we sought to determine the degree to which different scenes are more or less luminance equalized, and then correlated this with the change-detection results. To do this, we assessed the amount of rapid flicker present in each scene when the color and grey-scale versions of the scene were alternated at 15 Hz. Under these circumstances, equiluminant regions should not appear to flicker and instead should look less saturated (for review see Livingstone and Hubel 1988). An informal review of the flicking scenes verified that many regions in each scene did not flicker, although some did. Further, scenes did vary with regard to the proportion of each that had strong flicker (the strength of the flicker also appeared to vary). Therefore, we asked a total of seven judges (including the three authors) to rate each of the 32 scenes for flicker (both for grey-scale ^ color transients, and for luminance transients) on a 7-point scale ranging from ``no flicker'' (rating of 1) to ``maximum flicker'' (rating of 7). The mean between-judge rating correlation was 0.500, and all judges except one were significantly correlated with all other judges. Without the judge who showed non-significant correlations (4 of 7 of this judge's correlations were non-significant), the mean intercorrelation was Averaged ratings for each of the 32 images from the remaining six judges were uncorrelated with change detection for color changes for each of these images (r ˆ 0:045), and were also uncorrelated with the difference in change detection between color changes and no-transient changes for each scene (r ˆ 0:052). Change detection for color changes was correlated with change detection without transients (r ˆ 0:579, p 5 0:01), supporting the validity of the change-detection scores for each scene. (Note that this comparison contrasts performance across different subjects because a given scene was in the color-change condition for some subjects and in the no-transient condition for others.) These results suggest that luminance changes were not solely responsible for reduced change detection in experiments 1 and 2, and reinforce the hypothesis that a relatively broad category of transients can induce CB. This conclusion is further reinforced by the fact that the luminance changes did not overall produce more CB (when the results of experiment 3 are compared with those of experiments 1 and 2) than the color changes, despite the fact that the luminance changes produced much stronger flickering in the control experiment (judges rated the luminance-induced flicker to be stronger than the color-induced flicker of 31 or the 32 scenes). 6 General discussion In the three experiments we consistently observed that luminance and color transients resulted in a 26% ^ 39% reduction in change detection relative to a no-transient baseline. In experiment 1, a color onset reduced change detection by 33% relative to a no-transient baseline, whereas a flicker reduced change detection by 51% ^ 54%. Similarly, in experiment 2, color onsets and offsets reduced change detection by 30% ^ 39% relative to the no-transient baseline. Finally, in experiment 3, luminance transients produced a similar effect, reducing change detection by 26% ^ 34% relative to the no-transient baseline. These findings clearly demonstrate that substantial CB can be induced by a broad range of transients, even by those that do not involve the appearance of new objects,

12 1676 J G Arrington, D T Levin, D A Varakin or changes to object forms. However, it is important to note that the transients we tested were not as powerful as flickers in masking changes. This stands in contrast to the results of Turatto et al (2003), who found that contrast inversion and flicker caused equivalent levels of CB. There are at least two possible reasons for this difference. First, contrast inversion might be more powerful than either luminance or color transients in producing CB. Alternatively, the difference might be accounted for by differences in stimulus presentation between the studies. Turatto et al's contrast inversions actually introduced two transients before each change. In their paradigm, the contrast-inverted image replaced each flicker such that the original image was replaced by a contrastinverted image with no change for 100 ms, which in turn was replaced by a changed image with normal contrast relations. Thus, each change was associated with two contrast reversals (positive! negative and negative! positive). This is similar to the flicker paradigm in which each change is preceded by a disappearance and occurs simultaneously with a re-appearance. In contrast, the method we chose involved only a single transient, simultaneous with the change. Our single transients might have produced less CB than Turatto et al's two transients for two reasons. The most intuitive hypothesis is that two transients summate and together are more powerful than a single transient. Another possibility is that the transients are most powerful in blocking change detection when they precede the change by 80 ^ 100 ms. This might occur if awareness of the transient momentarily blocks an attentional bottleneck while the change might have been perceived or elaborated upon. Accordingly, the offset transient in the flicker paradigm, or the positive ^ negative transient in Turatto et al's experiment, would be the primary cause of CB. Recent findings do suggest that a pre-change transient [in this case the onset of new objects similar to O'Regan et al's (1999) mudsplash paradigm] can be powerful when subjects must identify the change by indicating the relative orientation of a rotated Gabor patch but not when they only locate the change (Watanabe 2003). However, even in this case, the transient is not more powerful than one simultaneous to the change and it is not clear whether the current paradigm is more comparable to Watanabe's location condition or to change-identification conditions. More generally, these results demonstrate that any of a broad category of transients is sufficient to induce CB. Accordingly, sudden appearances of objects, or transformations of form, are not unique in producing transients sufficient to block change detection. Most likely, any transient that either leads attention away from a change transient, or that reduces its salience, will be effective, and researchers have recently speculated that any of a variety of salient changes to objects may attract attention (Cole et al 2005). The ability of a relatively broad category of transients to capture attention may have been enhanced by the natural scene stimuli and changes we used. These required subjects to focus their attention on a relatively broad range of subtle form, texture, and orientation channels that may overlap with the color channels that would be activated by sudden changes in the presence of color information. Thus, the more general debate over the relationship between attentional focus and attentional capture may have another dimensionöthe breadth of attention across features that a given task setting requires. This kind of variation is particularly salient in natural change-detection tasks for which the target is not pre-specified on a given dimension but rather is a change to many of the dimensions of natural variation in a scene. Acknowledgments. We would like to thank Bonnie Angelone for helpful comments on earlier versions of this manuscript, along with Triniti Anderson, Kate Barker, and Jessica O'Dea for assisting in the running and scoring of the experiments. This research was supported by NSF grant #SES to DTL.

13 Color and luminance changes 1677 References Beck M R, Levin D T, 2003 ``The role of representational volatility in recognizing pre- and postchange objects'' Perception & Psychophysics ^468 Blackmore S J, Brelstaff G, Nelson K, Troscianko T, 1995 ``Is the richness of our visual world an illusion? Transsaccadic memory for complex scenes'' Perception ^ 1081 Cole G G, Kentridge R W, Heywood C A, 2005 ``Object onset and parvocellular guidance of attentional allocation'' Psychological Science ^ 274 Fernandez-Duque D, Thornton I, 2000 ``Change detection without awareness: Do explicit reports underestimate the representation of change in the visual system?'' Visual Cognition ^ 344 Fernandez-Duque D, Thornton I, 2003 ``Explicit mechanisms do not account for implicit localization and identification of change: An empirical reply to Mitroff et al (2002)'' Journal of Experimental Psychology: Human Perception and Performance ^ 858 Folk C L, Remington R W, Johnston J C, 1992 ``Involuntary covert orienting is contingent on attentional control settings'' Journal of Experimental Psychology: Human Perception and Performance ^ 1044 Gellatly A, Cole G, Blurton A, 1999 ``Do equiluminant object onsets capture visual attention?'' Journal of Experimental Psychology: Human Perception and Performance ^ 1624 Grimes J, 1996 ``On the failure to detect changes in scenes across saccades'', in Vancouver Studies in Cognitive Science. Volume 2: Perception Ed. K Atkins (New York: Oxford University Press) pp 89 ^ 110 Hollingworth A, Henderson J M, 2000 ``Semantic informativeness mediates the detection of changes in natural scenes'' Visual Cognition ^ 235 Jonides J, Yantis S, 1988 ``Uniqueness of abrupt visual onset in capturing attention'' Perception & Psychophysics ^354 Kemp R, Pike G, White P, Musselman A, 1996 ``Perception and recognition of normal and negative faces: the role of shape from shading and pigmentation cues'' Perception ^ 52 Levin D T, Simons D J, 1997 ``Failure to detect changes to attended objects in motion pictures'' Psychonomic Bulletin and Review ^ 506 Livingstone M, Hubel D, 1988 ``Segregation of form, color, movement and depth: Anatomy, physiology and perception'' Science ^ 749 Lu S, Zhou K, 2005 ``Stimulus driven attentional capture by equiluminant color change'' Psychonomic Bulletin and Review ^ 572 McConkie G W, Zola D, 1979 ``Is visual information integrated across successive fixations in reading?'' Perception & Psychophysics ^ 224 Mondy S, Coltheart V, 2000 ``Detection and identification of change in naturalistic scenes'' Visual Cognition ^ 296 O'Regan J K, Rensink R A, Clark J J, 1999 ``Change blindness as a result of `mudsplashes''' Nature Posner M I, Snyder C R R, Davidson B J, 1980 `Àttention and the detection of signals'' Journal of Experimental Psychology: General ^ 174 Poynton C, 1998 ``Frequently asked questions about color'', available at ColorFAQ.html Rensink R A, 2002 ``Change detection'' Annual Review of Psychology ^ 277 Rensink R A, O'Regan J K, Clark J J, 1997 ``To see or not to see: The need for attention to perceive changes in scenes'' Psychological Science 8 368^373 Shapley R, 1995 ``Parallel neural pathways and visual function'', in The Cognitive Neurosciences Ed. M S Gazzaniga (Cambridge, MA: MIT Press) pp 315 ^ 324 Simons D J, 2000 ``Attentional capture and inattentional blindness'' Trends in Cognitive Sciences ^ 155 Simons D J, Chabris C F, 1999 ``Gorillas in our midst: Sustained inattentional blindness for dynamic events'' Perception ^1074 Simons D J, Franconeri S L, Reimer R L, 2000 ``Change blindness in the absence of a visual disruption'' Perception ^ 1154 Simons D J, Levin D T, 1997 ``Change blindness'' Trends in Cognitive Sciences ^ 267 Simons D J, Levin D T, 1998 ``Failure to detect changes to people during a real-world interaction'' Psychonomic Bulletin and Review ^ 649 Simons D J, Levin D T, 2003 ``What makes change blindness interesting?'', in The Psychology of Learning and Motivation volume 42, Eds D E Irwin, B H Ross (San Diego, CA: Academic Press) pp 295 ^ 322 Snowden R J, 2002 ``Visual attention to color: Parvocellular guidance of attentional resources?'' Psychological Science ^ 184

14 1678 J G Arrington, D T Levin, D A Varakin Theeuwes J, 1995 `Àbrupt luminance change pops out; abrupt color change does not'' Perception & Psychophysics ^ 644 Turatto M, Bettella S, Umilta C, Bridgeman B, 2003 ``Perceptual conditions necessary to induce change blindness'' Visual Cognition ^ 255 Watanabe K, 2003 ``Differential effect of distractor timing on localizing versus identifying changes'' Cognition ^ 257 Wayand J W, Levin D T, Varakin D A, 2005 ``Inattentional blindness for a noxious multimodal stimulus'' American Journal of Psychology ^ 352 ß 2006 a Pion publication

15 ISSN (print) ISSN (electronic) Conditions of use. This article may be downloaded from the Perception website for personal research by members of subscribing organisations. Authors are entitled to distribute their own article (in printed form or by ) to up to 50 people. This PDF may not be placed on any website (or other online distribution system) without permission of the publisher.