VISUAL COGNITION, 2006, 14 (4/5/6/7/8), 704 715 Evidence for a systematic component within scan paths in visual search Iain D. Gilchrist Department of Psychology, University of Bristol, UK Monika Harvey Department of Psychology, University of Glasgow, UK We present evidence that scan paths in visual search can include a systematic component. The task for subjects in the experiment was to search for a target that was either present or absent. With regular grid-like displays, participants generated more horizontal saccades than vertical saccades. Disruption of the grid structure in the display modulated but did not eliminate the systematic component. This is consistent with the scan path being partly determined by a cognitive strategy. We discuss the implications of this finding for studies that use refixation to investigate memory mechanisms in visual search. Monitoring eye movements during visual search has become an important method for investigating the processes that mediate between display onset and the manual response (e.g., Findlay, 1997; Hooge & Erkelens, 1996, 1999; Zelinsky & Sheinberg, 1997). Models of visual search and particularly ones that include the programming of saccades, have tended to focus on bottom up salience driven mechanisms for saccade allocation to items in the display (e.g., Itti & Koch, 2000). At the centre of most of these models is a salience map of one form or another. The salience of items is determined by the strength of the visual input to the system and by the characteristics of the target. So, for example if the target is a green vertical bar then green items and vertical items will have enhanced salience. This is the mechanism by which attention (Wolfe, 1994; Wolfe, Cave, & Franzel, 1989) or saccades (Findlay & Walker, 1999; Itti & Koch, 2000; Wolfe & Gancarz, 1996) are deployed to items that are more similar to the target. Over and above these Please address all correspondence to I. D. Gilchrist, Department of Psychology, University of Bristol, Bristol BS8 1TN, UK. E-mail: [email protected] This work was supported by a grant from the EPSRC UK (grant no: GR/M37295). We would like to thank Casimir Ludwig for help with data collection. http://www.psypress.com/viscog # 2006 Psychology Press Ltd DOI: 10.1080/13506280500193719 SCAN PATHS IN VISUAL SEARCH 705 salience-based processes it is assumed that the location to be fixated next is determined by a random process (Itti & Koch, 2000; Scinto, Pillalamarri, & Karsh, 1986). However, four decades ago Williams (1966) suggested that there was often a bias in the direction of saccades in search and that these biases reflected a systematic process rather than the action of a random process (see also Norton & Stark, 1971a, 1971b; Yarbus, 1967). This observation is of course consistent with introspection. When search becomes difficult we tend to start in one place and work systematically through the display. In this paper we have used Williams’ (1966) term ‘‘systematic’’ to describe this nonrandom component in the scan path. Although there have been qualitative reports of a systematic component in scanning (e.g., Hooge & Erkelens, 1996), it has not been studied quantitively. Indeed early attempts to quantify scan paths in visual search lead to the conclusion that randomwalk models best characterized the allocation of fixations in search (Scinto et al., 1986). More recently, Motter and Belky (1998) investigated the extent to which the previous saccade influenced the characteristics of the subsequent saccade. They found no evidence for either the length or the direction of the previous saccade having a large influence on the subsequent saccade, except for a slight bias for the next saccade to avoid the area just crossed by the saccade. One of the reasons for the lack of evidence for systematic scanning in visual search may be that although scanning could be purely systematic, such a process will often operate alongside other saliencebased processes to determine which item is fixated next. If systematic scanning is not the sole factor determining the structure of the scan path it may not be easy to detect its presence. In this paper we report an experiment in which we record eye movements in visual search. The purpose of this report is to (1) describe a method by which systematic patterns in scan path structure can be identified, (2) demonstrate that systematic scanning does occur in visual search, and (3) show that the extent of this behaviour is modulated by, but relatively robust to, substantial disruption of display structure. One of the difficulties of detecting the presence of systematic scanning behaviour is developing methods by which it can be detected both across trials and across participants and then analysed using statistical techniques. Following Williams’ (1966) claim that there is often a directionality to successive fixations, in the current experiment we analysed the frequency of saccade directions. Analysis of saccade direction has the potential to reveal systematic patterns, even when they are combined with other mechanisms to determine the final scan path. In the current experiment, across three conditions, we manipulated the regularity of the display while keeping display size and the display elements constant and analysed the frequency of saccades in each direction across conditions. 706 GILCHRIST AND HARVEY EXPERIMENT In all three conditions the target was an upright triangle, and the distractor items were downward pointing triangles and leftward pointing triangles. In Condition 1 the items were completely regular, less regular in Condition 2, and even less so in Condition 3. Method Participants. Twelve adults (9 female and 3 male, age range 1945 years) took part in the study. All were from the University of Bristol and were paid, or received course credit, for their time. Design and procedure. The experiment was a repeated measures design, with the order of condition presentation presented in separate blocks and counterbalanced across the participants. Viewing distance was 57 cm. The participants were instructed to make a present/absent response using a button box. At the beginning of each trial, a small white disc with a black dot in its centre appeared in the middle of the screen. Online compensation for any spatial offset in the calculated eye position was made at this point before each trial. The display was then presented until a manual response was made and, when appropriate, this was followed by an error message presented centrally on the screen. Participants were given verbal instructions describing the task and asked to respond as quickly and accurately as possible. Displays were presented on a 17-inch SVGA monitor with 800 /600 pixel resolution. A chinrest was used to minimise head movements. In all conditions the target was an upright triangle and the distractor items were triangles pointing downwards and pointing to the left (see Figure 1), display size was 25 items. In trials where the target was not present, the target was replaced with one of the distractor types. Each of the three blocks contained 98 trials. Participants were given a break of approximately 3 minutes between blocks of trials. In Condition 1 the 25 items were placed randomly on the junctions of an imaginary 5 /5 grid resulting in a spatially regular display. In Condition 2 the same 25 display items were placed randomly on the junctions of an imaginary 6 /6 grid leaving 12 randomly selected locations blank in each trial. And in Condition 3, the display items were placed randomly on the junctions of an imaginary 7 /7 grid, leaving 24 randomly selected locations blank in each trial. Across all three conditions the overall display size was kept constant (12 /12 deg). As a result, Condition 1 was the most spatially structured Figure 1. Example displays: Condition 1 (upper panel), Condition 2 (middle panel), and Condition 3 (lower panel). 707 708 GILCHRIST AND HARVEY display and Condition 3 was the least structured display. Example displays can be seen in Figure 1. Eye movement recording and analysis. Two dimensional, binocular eye movements were recorded using an SMI Eye-Link eyetracker (SensoMotoric Instruments GmbH, Berlin, Germany). The Eye-Link system uses an infrared video technique sampling at 250 Hz, and features a head movement compensation mechanism. Displays were presented on one PC (subject PC), while a second PC (operator PC) recorded the eye position data online. Each block of trials was preceded by a nine-point calibration and validation procedure. The eye position data were analysed off line by an automatic saccade detection procedure. For each participant, the data were analysed only from the eye that produced the best spatial resolution, which in this experiment was typically 0.20 deg. A fixation was defined as having ended when the eye velocity exceeded 30 deg/s. A fixation began after the velocity fell below this value for five successive samples (20 ms). This rather stringent criterion excludes the interval of ocular instability just after the saccade and so leads to a more accurate calculation of fixation location which is an important measure in the current experiment. However, as a result the fixation durations reported in this paper may be shorter (by approximately 15 ms) than those typically reported elsewhere. The absolute direction of all the saccades larger than 1 degree was analysed. Saccades were coded for direction in degrees with saccades in a vertical direction coded as zero. For each participant, in each condition, the frequency of saccades in each direction in 10 degree frequency bins was calculated. As the total number of saccades was large, even for a single participant, these data could be analysed using standard parametric tests. The percentage of fixations outside the display area was 3.3% in Condition 1, 4.5% in Condition 2, and 6.1% in Condition 3. These fixations were excluded for subsequent analyses. Results Errors. The percentages of manual response error trials per condition were 6.72%, 7.14%, and 6.38%, for Conditions 1, 2, and 3 respectively. Most errors occurred on target-present trials (13.3%) rather than target-absent trials (0.68%). Error trials were excluded from later analyses. Reaction times. Mean reaction times in the three conditions are shown in Table 1. A repeated measures ANOVA was carried out on these data. Reaction times were slower for target-absent (4665 ms) than for targetpresent (2549 ms), F (1, 11) /63.7, p B/ .001; but there was no reliable effect SCAN PATHS IN VISUAL SEARCH 709 TABLE 1 The mean reaction times (in ms), fixation duration (in ms), and number of fixations in the experiment; values are given for target-present and target-absent trials by condition, with standard deviations in parentheses Condition 1 Measure Condition 2 Condition 3 Present Absent Present Absent Present Absent Mean reaction time (ms) SD 2443 (755) 4596 (1663) 2752 (849) 4723 (1626) 2453 (650) 4676 (1604) Mean fixation duration (ms) SD 225 (29) 217 (23) 221 (22) 213 (20) 221 (21) 214 (19) Mean number of fixations SD 10.7 (3.0) 18.5 (6.2) 11.6 (3.2) 18.3 (5.9) 11.0 (2.8) 19.0 (6.10) of Condition, F (2, 22) B/1. There was also no significant interaction between condition and target presence, F (2, 22) B/1. Inspection of Table 1 revealed that numerically there were only very small differences between the conditions. Fixation duration and fixation number. The total number of fixations examined in each condition were 17,196 in Condition 1, 17,619 in Condition 2, and 17,635 in Condition 3. Mean fixation duration and the number of fixations per trial are shown in Table 1. A repeated measures ANOVAs revealed a significant effect of target presence on fixation duration, with fixations in target present trials being longer (222 ms) than fixations in target absent trials (215 ms), F (1, 11) /12.7, p B/ .01. However, there was no significant effect of condition, F (2, 22) /2.15, p/ .140, nor a significant interaction, F (2, 22) B/1. For the number of fixations per trial there was again a significant effect of target presence in that there were, on average, more fixations in the target absent trials (18.6 fixations) than the target present trials (11.1 fixations), F (1, 11) /58.9, p B/ .001. There were no reliable effects of condition, F (2, 22) B/1. The interaction did not reach significance but was marginal, F(2, 22) /2.90, ns. p / .08. For fixation duration and fixation number, there was no evidence for reliable differences between conditions. Saccade direction. The distribution of saccade directions in the three conditions is shown in Figure 2. A repeated measures ANOVA was carried out on saccade direction frequency. There was no significant effect of condition, F (2, 22) B/1. However, there was a significant effect of direction, F (35, 385) /15.5, 710 18 18 18 12 12 12 6 6 6 0 0 0 -6 -6 -6 -12 -12 -12 -18 -18 a -12 -6 0 6 12 18 -18 -18 b -12 -6 0 6 12 18 -18 -18 -12 -6 0 6 12 c Figure 2. The distribution of the angle of movement of each saccade in Condition 1 (panel a), Condition 2 (panel b), and Condition 3 (panel c). 18 SCAN PATHS IN VISUAL SEARCH 711 p B/.001, indicating a reliable increase in the number of saccades generated in a specific direction consistent with a systematic component in the scan paths. In addition there was a significant interaction, F (70, 770) /5.75, p B/ .001. Pairwise comparisons between each of the three conditions showed a significant interaction in all cases: Conditions 1 and 2, F (35, 385) /4.56, p B/ .001; Conditions 2 and 3, F (35, 385) /3.25, p B/ .001; Conditions 1 and 3, F (35, 385) /7.73, p B/ .001. These interactions reflect a decreasing asymmetry in the saccade direction distributions from Conditions 1 to 3. There was evidence for systematic scanning in all three conditions as reflected in the main effect of direction. The systematic scanning effect took the form of more horizontal saccades than other directions. In addition this effect was strongest for the more structured displays. These effects on saccade direction were not apparent from simple inspection of single scan paths, and it certainly was not possible to differentiate the extent of these effects between conditions by inspection alone. Discussion In the current experiment we found strong evidence for systematic scanning in visual search. The method introduced allowed the detection of properties of the scanning that were consistent across participants and across conditions. The extent of systematic scanning was modulated by display structure: As the display became less regular across conditions so the asymmetries in the directions of the saccades produced was reduced. However, there was even evidence for such asymmetries in Condition 3 where the regularity of the display had been significantly disrupted. One possibility is that the asymmetries in the directions of the saccades produced here simply reflect low-level oculomotor biases. For example, even the organization of oculomotor muscles may lead to biases in the number of saccades in each direction. However, here we have shown that the extent of these asymmetries differs across different display conditions. If the effect was a result of low-level oculomotor factors we would expect it to be constant across condition. In a comparison between the conditions, we found no reliable differences between reaction time and number of fixations. This suggests that task difficulty was relatively well matched across conditions. However, as the distribution of saccade direction shows, the manipulation of display regularity had a systematic and reliable effect on the nature of the scanning. The experiment also allows an important additional conclusion to be drawn. Although systematic scanning was modulated by display regularity, the results from Conditions 2 and 3 confirm that systematic scanning occurs even for displays that are not perfectly regular. Indeed the results suggest that these effects are relatively robust to quite large 712 GILCHRIST AND HARVEY disruptions of the regularity of the structure as can be seen in Condition 3. If the regularity of the display structure were reduced further the presence of any detectable systematic behaviour may be further reduced and even disappear. The current experiment did not investigate this change in detail. Instead it demonstrated that systematic scanning does not depend solely on a completely regular display and is not dependent solely on task difficulty. However, most naturalist visual scenes do contain an extensive complex spatial structure (Marr, 1982). In these more visually complex conditions it may be that this structure shapes systematic scanning. The form that the systematic scanning took in this experiment was almost certainly a function of the type of structure that was imposed on the displays: specifically the display items were placed randomly on the intersections of a grid. However, previous work by Hooge and Erkelens (1996) has shown that participants will systematically scan around circular displays, suggesting that systematic scanning is not a behaviour restricted only to grid based displays. Memory processes have an influence on search behaviour at a number of levels (Shore & Klein, 2000). However, one contentious issue is the extent to which visual search relies on memory mechanisms to prevent items that have been inspected from being reinspected. Estimates of the capacity for this type of memory have ranged from no memory at all (Horowitz & Wolfe, 1998) to a limited memory capacity (Gilchrist & Harvey, 2000), and to a more extensive memory capacity (Peterson, Kramer, Wang, Irwin, & McCarley, 2001). One approach to this question has been to record eye movements in order to get a more direct measure of where attention is allocated at any one time. The key measure here has been the extent of refixation of distractors. The logic is that when no refixation occurs then the items have been remembered. A number of groups have modelled this kind of data to obtain an estimate of the capacity of the memory store (Gilchrist & Harvey, 2000; Peterson et al., 2001). This leads to two key, interrelated, questions. The first is what does it mean when an item is refixated? And the second is, what does is mean when an item is not refixated? The answer to neither of these questions is straightforward. What does it mean when an item is refixated? Refixating a distractor item in visual search appears to suggest that the subject has forgotten that the item has been visited and so is returning to reinspect it. However, as Gilchrist and Harvey (2000) pointed out, refixations can also occur because the participant moved away from the fixated item before the processing of that item was completed. The refixation in this case simply reflects a return to the item to complete processing. Indeed such deadline limiting of fixation duration is part of the influential model of saccade control developed by Henderson (1992). What does it mean when an item is not fixated? One possibility is that visual processing in the periphery has ruled out that item as a possible target. This process of SCAN PATHS IN VISUAL SEARCH 713 visual guidance is an important determinant of saccadic selection in search (e.g., Findlay, 1997). However, a number of researches in this field have ruled out guidance by using display items that are very small, or very similar to that target (e.g., Peterson et al., 2001), so that it is impossible to distinguish the target from distractors without fixating. It would appear then that if the stimuli in the search task are correctly selected, an item not being fixated is evidence for memory. However, the story may not be that simple. The evidence presented here suggests that saccades in visual search can be systematic. If the saccades generated by participants are not random, but instead have a systematic component, then this itself can reduce refixations. In the most extreme case, where participants follow a fixed route thorough the display no refixations will occur. The lack of refixations would not be a product of remembering specifically which items had been visited but instead would reflect the consequence of following a predictable route. If such a mechanism does influence scan paths then a lack of refixations in a task may not be a hallmark of memory in search. In turn this suggests that when the display has a structure that can drive systematic scanning it is possible that systematic scanning can substitute for memory. If participants follow broadly the same scan path on each trial this would allow them to eliminate part of the display simply because it occurred earlier in the search. In the current experiment there is a decrease in the extent of strategic scanning across conditions. If the above argument is correct, then we would expect a concurrent increase in task difficulty because participants were less able to rely on systematic scanning to support efficient search. This increase in task difficulty should be reflected in an increase in the number of fixations and an increase in the overall response time: We found neither in the current experiment. There are a number of possible explanations for this apparent anomaly. Various authors have suggested that memory capacity in search is quite large (e.g., Peterson et al., 2001); however, the deployment of this memory might in itself require effort (Gilchrist, North, & Hood, 2001). In the current experiment what may have occurred is a tradeoff between memory deployment and systematic behaviour resulting in approximately equivalent overall search performance. Another possibly is that, in the less regular displays, participants were still engaging in the same amount of systematic scanning but that the particular form of scanning adopted varied either across subjects or even trial-to-trial. If the form of the systematic scanning is inconsistent then the method presented in this paper is unable to detect its presence*instead the method presented here detects systematic scanning that is consistent across trials and participants. Further work would clearly be needed to disentangle these possibilities. However, whatever the explanation, the current experiment demonstrates the presence of systematic scanning in search. The presence of systematic scanning does not rule out memory processes as being important in search. However, what 714 GILCHRIST AND HARVEY is clear is that a number of mechanisms appear to structure scan paths in more difficult visual search. Understanding the relationships between these mechanisms and developing methods to detect and quantify their effects is an important step in understanding the search process. As a step towards this, the present paper presents a method for detecting and statistically analysing systematic components in scan paths. REFERENCES Findlay, J. M. (1997). Saccade target selection during visual search. Vision Research , 37 , 617 631. Findlay, J. M., & Walker, R. (1999). A model of saccade generation based on parallel processing and competitive inhibition. Behavioral and Brain Science, 22 , 661 721. Gilchrist, I. D., & Harvey, M. (2000). Refixation frequency and memory mechanisms in visual search. Current Biology, 10 , 1209 1212. Gilchrist, I. D., North, A., & Hood, B. (2001). Is visual search really like foraging? Perception , 30 , 1459 1464. Henderson, J. M. (1992). Visual attention and eye movement control during reading and picture viewing. In K. Rayner (Ed.), Eye movements and visual cognition (pp. 261 283). Berlin: Springer-Verlag. Hooge, L. T. C., & Erkelens, C. J. (1996). Control of fixation duration in a simple search task. Perception and Psychophysics, 58 , 969 976. Hooge, I. T. C., & Erkelens, C. J. (1999). Peripheral vision and oculomotor control during visual search. Vision Research , 39 , 1567 1575. Horowitz, T. S., & Wolfe, J. M. (1998). Visual search has no memory. Nature, 394 , 575 577. Itti, L., & Koch, C. (2000). A saliency-based search mechanism for overt and covert shifts of visual attention. Vision Research , 40 , 1489 1506. Marr, D. (1982). Vision . San Francisco: W. H. Freeman. Motter, B. C., & Belky, E. J. (1998). The guidance of eye movements during active visual search. Vision Research , 38 , 1805 1815. Norton, D., & Stark, L. (1971a). Eye movements in visual perception. Scientific American , 224 , 34 43. Norton, D., & Stark, L. (1971b). Scanpaths in saccadic eye movements while viewing and recognising patterns. Vision Research , 11 , 929 942. Peterson, M. S., Kramer, A. F., Wang, R. X. F., Irwin, D. E., & McCarley, J. S. (2001). Visual search has memory. Psychological Science, 12 , 287 292. Scinto, L. F. M., Pillalamarri, R., & Karsh, R. (1986). Cognitive strategies for visual search. Acta Psychologica , 62 , 263 292. Shore, D. I., & Klein, R. M. (2000). On the manifestations of memory in visual search. Spatial Vision , 14 , 59 75. Williams, L. G. (1966). The effect of target specification on objects fixated during visual search. Perception and Psychophysics, 1 , 315 318. Wolfe, J. M. (1994). Guided Search 2.0: A revised model of visual search. Psychonomic Bulletin and Review, 1 , 202 238. Wolfe, J. M., Cave, K. R., & Franzel, S. L. (1989). Guided Search: An alternative to the feature integration model for visual search. Journal of Experimental Psychology: Human Perception and Performance, 15 , 419 433. SCAN PATHS IN VISUAL SEARCH 715 Wolfe, J. M., & Gancarz, G. (1996). Guided Search 3.0: A model of visual search catches up with Jay Enoch 40 years later. In V. Lakshminarayanan (Ed.), Basic and clinical applications of vision science (pp. 189 192). Berkeley, CA: Kluwer Academic Publishers. Yarbus, A. L. (1967). Eye movements and vision . New York: Plenum Press. Zelinsky, G. J., & Sheinberg, D. L. (1997). Eye movements during parallel-serial visual search. Journal of Experimental Psychology: Human Perception and Performance, 23 , 244 262.
© Copyright 2026 Paperzz