This experiment examined the effect that different set-sizes and proximity distances had on a feature integration visual search task. Different set-sizes were used as there have been mixed findings as to whether the amount of distractors have an effect on participants reaction times when locating a target. Proximity explored whether large and small display sizes had an effect on visual ‘pop-out’. It was concluded that reaction times were slowest in smaller set-sizes and distant proximity conditions and this was supported with asignificant interaction. These findings are similar to experiments carried out by Schubo, Schroger and Meinecke (2004) on set-size manipulation, although the reasons for the results remain inconclusive. The results support Wolfe’s (1998) theory that it may make more sense to talk about different degrees of efficiency in visual search tasks as opposed to the distinction between parallel and serial processing.
Analysis of one’s perceptual experiences and how these are broken down into more elementary sensations proves a difficult task, due to the subjective activity of perceptual attention (Treisman & Gelade, 1980). Physiological evidence suggests that specialized populations of receptors respond selectively to individual properties mapped to different parts of the brain, which are then analyzed at an early stage in visual processing (Zeki, 1976). In the present paper, the role that Treisman’s (1980) feature integration theory (FIT), set size manipulation, and also large and small proximity displays play in visual attention using target ‘pop-out’ is investigated. There have also been ambiguous results in the sensitivity of the distribution of stimuli (Hoffman & Sebald, 2005). These issues will be discussed in the following paper.
According to Treisman’s (1980) feature integration theory, visual display is processed into two successive stages. Firstly, one’s field of vision is available to be registered by the brain via parallel processing which consists of a set of spatio-topically organized “maps” of the visual field, each coding the presence of a particular “feature” e.g. color, edges, orientation, location and movement (Duncan & Humphreys, 1989). Secondly, serial processing combines all the separate features of a particular object at a particular location together so that the object can be synthesised and identified (Treisman & Gelade, 1980). A study by Posner (1999) supports this theory that, rather than precuing the location there is a far better benefit precuing the modality of the stimulus as early processing picks up on separate specific features of an object, then registers individual specific features as a whole during late visual processing (Treisman & Gelade, 1980).
Contrastingly, in studies carried out by Duncan and Humphries (1989), non-target homogeneity was not always necessarily the case in creating faster parallel searches. Duncan and Humphries (1989) found that search efficiency decreased with (a) increasing similarity between targets and non-targets (which they called T-N similarity) and (b) decreasing similarity between non-targets themselves (N-N similarity), both creating overlaps with opposing interactions (Duncan & Humphreys, 1989). Some preliminary results reported by Treisman (1988) suggested that non-target heterogeneity on irrelevant dimensions had little effect on the target. Incidentally, by varying the number of non-target colours, Farmer and Taylor (1980) showed that this was not necessarily the case and that there were detrimental effects of non-target heterogeneity.
Additionally, research by Hoffman & Sebald (2005) produced convincing evidence for a sensitivity of the distribution of stimuli. If different stimulus targets appeared in different locations with unequal probabilities, search performance closely adapted to context-dependent distributions of target locations (Hoffman & Sebald, 2005). Therefore, in a global search of target stimuli, contextual-cuing could lead to certain locations where the target was most likely to be found (Hoffman & Sebald, 2005). Hoffman & Sebald (2005) had participants searching for target-letters amongst different distractor levels. Thusly, the stimuli were evenly distributed in a virtual circle, which formed an invariant homogeneous layout in each trial (Hoffman & Sebald, 2005). Subsequently, two distractors were frequently positioned next to the target, and another two distractors were rarely positioned next to the target. As a result, participants were more likely to detect targets with frequent stimuli than targets with rare stimuli (Hoffman & Sebald, 2005).
Subsequently, Poder (2005) looked at the crowding effect and colour ‘pop-out’ priming. Poder’s (2005) studies had shown that almost equal facilitation of crowded letters could be observed with differently coloured targets and non-targets. The most reasonable conclusion seemed to be that one’s attention was attracted to the location of saliently coloured stimuli (Poder, 2005). Also, spatial extent of crowding depended on the eccentricity of stimuli in the visual field. Alas, when the distance between target and non-target stimuli was larger than approximately 0.5E (retinal eccentricity), adverse interaction disappeared (Poder, 2005). Contrastingly, Bouma, Toet and Levi (1992) inferred that non-targets would only affect target identification if they fell within a spatial region known as the interference zone. Moreover, Bouma et al’s., studies supported that interference zones grew with a target’s eccentricity but were independent of target size. Subsequently, one’s ability to identify objects in one’s peripheral field of vision compared to the fovea is vulnerable to the presence of nearby visual clutter, a phenomenon known as ‘visual crowding’ (Levi, 2008). This finding holds across “low-level” stimulus dimensions including; contrast polarity, colour, orientation, spatial frequency, direction and speed (Triesman, 1988).
Furthermore, Schubo, Schroger and Meinecke (2004) noted a decrease in RT’s when differentiating the proximity of stimuli. Thusly, Schubo et al’s., (2004) experiment recorded participants performance first decreasing with an increase in proximity before improving again when proximity was further increased. Consequently, in all studies based on the effect of proximity and pop-out detection, a performance decrease was first observed (Schubo, et al., 2004). Generally, a sharp distinction between parallel and serial visual search is no longer maintained (Wolfe, 1998). Thusly, instead of exclusively referring to the distinction between parallel and serial processing, it may make more sense to talk about different degrees of efficiency in visual search tasks Wolfe (1998).Moreover, the response pattern found for small set-sizes (performance decrease) and large set-sizes (performance increase) correspond to what is usually found in visual search and in texture segmentation tasks (Eckstein, Thomas, Palme, Shimozaki, 2000).
The current research question addressed some of the uncertainty based around the relationship between stimuli proximity and set-size. The aim of this research was to investigate if attention was facilitated towards visual ‘pop-out’ regardless of set size and proximity manipulation. One of the aims was to keep the stimuli order as random as possible to prevent a contextually dependent bias. The null hypothesis was that there would be no effect of the number of distractors on the response times and that the RTs for close proximity’s would be longer than the RTs in the distant proximity’s.
This experiment used a 2×2 repeated-measures design. The first independent variable was set-size manipulation. There were two conditions; Level 1- Black target circle amongst two shades of 8 grey non-target circles displayed on a 3×3 grid. Level 2- Black target circle amongst two shades of 24 grey non-target circles displayed on a 5×5 grid. The second independent variable was proximity, which was distance between target and non-target stimuli. This was manipulated by producing a grid of elements on two scales: small (level 1) and large (level 2). The (invisible) matrix grid was scaled to subtend a large or small visual angle, both subtended the same visual angle of about 0.01 rad or 0.57°. Stimuli were 0.5 cm in all conditions therefore; changing the size of each grid but not changing the elements kept consistency. The distractors and target stimuli differed in chromaticity only. Participants were instructed to decide whether the stimulus arrays contained a target (target trials) or not (blank trials). The experiment lasted for a block of 160 trials with 40 used for each of the conditions with 20 target trials and 20 blank trials in each. The nature of the trial block was to gauge if the participants fully understood what was asked of them and this lasted for a block of 10 trials.
An opportunity sample of 21 Open University under-graduate psychology students participated in the research as part of their course requirement. Counterbalancing was also employed to prevent order effects such as fatigue, task familiarity and boredom.
The stimulus presentation was carried out using E-Prime. The stimuli were created in Microsoft Powerpoint with the bigger size sets measuring 300 x 300 pixels and the smaller size sets measuring 50 x 50 pixels. The three grey scale shades used were 35%, 50% and 100% using Powerpoint. The stimuli were displayed on an HP Compaq LA2006x (LED monitor – 20″)monitor that had a screen resolution of 1600 x 900 at 60 Hz and a refresh rate of 76hz (vertical) and 83khz (horizontal).
Participants were placed 100 cm from a computer screen and the fixation point was at the center of their field of vision. Participants were told that they would see a series of images on the screen, preceded by a fixation cross. The participants were given instructions to look for a black filled circle amongst gray scale circles. The participants were instructed to press ‘z’ if a black circle target appeared on the screen, and press ‘m’ if no target appeared. The duration of trial block was set at ‘infinite’ until the participants had responded with a key push on the keyboard. The participants ran through a practice block before they felt comfortable to run the experiment. A pilot was run to ensure that the instructions were comprehensive enough for the participants and to make sure the experiment was ready to run. Instructions were tweaked and the numbers of trial blocks were increased as a result of the pilot run. Due to the nature and design of the experiment we could only test participants with corrected vision. This was cleared by ethics during project proposal stage’.
F (IV df, error df) = F-ratio, p=Sig Therefore (F(1,22) = 0.745, p = 0.707)
While there was no significant main effect of ‘Pop-out’ (F(1,22) = 0.707 p<0.707), there was a significant interaction between proximity and set-size on response times which was unexpected, with participants responding on average 25.5 ms slower for distant proximity and large set-size.
(F(1,22) = 6.103, p= 0.022
The results from this experiment reported no main effect for set-size and no main effect for proximity. However, there was a significant interaction between set-size and proximity. The response times for small and large set-size conditions were on average similar with (mean = 692.84) for large set-size and (mean = 696.34) for smaller set-size. Therefore, set sizes did not affect ‘pop-out’. Participants’ reaction times (RT’s) were, on average, 4.6 milliseconds (ms) faster for small set-size and close proximity; therefore visual ‘pop-out’ was most effective in these conditions. Surprisingly, participants’ reaction times for distant proximity were, on average 10.9 milliseconds (ms) faster than close proximity in the overall proximity condition. Whereas, for small set-sizes participants were faster, responding to close proximity (mean = 694.6) than distant proximity (mean = 698.1) which was 3.5 milliseconds (ms) faster on average but there was no main effect. Lastly, there was a significant interaction between large set-size and distant proximity, in which, on average, participants RTs were 25.5 milliseconds (ms) slower in the distant and large set-size condition. However, the reason for this interaction could be related to the well-known decline in visual acuity in the peripheral vision (Levi, 2008).
The results support Treisman’s (1988) feature integration theory, which states that, varying the amount of stimuli on the display does not affect response times. Also, the results fail to support the crowding effect, since response times were longer in the condition where the proximity between items was distant rather than close in the large set-size. As a result, different set-sizes might have induced different search strategies in second order (parallel) and first order (serial) processing. This, in turn, could have determined whether ‘pop-out’ occurred or not. The results in this paper are in stark contrast to Poder’s findings (2007) that, during a ‘pop-out’ orientation task, the type of processing was dependent on the number of elements presented in the stimulus display together with the target.However, again it might make more sense to speak of different degrees of efficiency (Wolfe, 1998).
Consequently, Meinecke & Donk (2002) concluded that three different processing modes could be observed in their experiments depending on proximity and in different set-sizes. Consequently, the three models proposed could give better insight towards the current results in this paper. Firstly, there is the single-element hypothesis, which occurs when display size is relatively small and a performance decrease is found when the set-size is increased therefore also increasing the retinal eccentricity of the target. Although there was no significance to our findings, there was a similar pattern in the small set sizes. Secondly, there is the spatial-integration hypothesis, which generally occurs when the display size is relatively large and there is a notable performance increase when density is increased and thus, retinal eccentricity is also increased. This is due to smaller inter-element distances and larger receptive fields outside the fovea (Meineck & Donk, 2002). Surprisingly, with our findings, there was a notable increase in RTs with larger proximity and smaller set size, which does not fully support the spatial-integration hypothesis. However, larger receptive fields are stronger in the peripheral regions than in the foveal region and spatial-integration could be an automatic byproduct when the stimulus is projected onto peripheral regions (Levi, 2008).
Contrastingly, Levi & Klien (1990) suggest decreased sampling density of retinal ganglion cells leads to greater intrinsic blur in the periphery, and this could perhaps be sufficient enough reason to expect longer RTs in more distant proximity. Our results showed slower RT’s in distant proximity but in the small set size.In accordance to this, this could be the reason that there was a significant interaction with set size and proximity due to greater intrinsic blur preventing preattentive ‘pop-out’ when switching from large to small proximity. Another of Meineck & Donk’s (2002) processing modes is the irregularity-detection hypothesis. This is the result of the target being randomly displayed on an otherwise regular background. Consequently, just like this paper, all the cells of the (invisible) matrix were filled with homogenous (shape gradient) elements and in target trials; the target again deviated from an otherwise full matrix grid displayed on a normal background making the target ‘pop-out’. There is a possibility that this could have influenced detection performance and thus, created faster RTs.
Furthermore, the interaction could have been due to visuo-spatial aspects when shifting attention from smaller set-sizes to larger set-sizes that cover more ground (Triesman, 1988). Apparently, in theory, the bigger the proximity and set-size, the more serial approach is used in a participants search performance. In turn, this would mean bigger saccades when looking for the target stimuli, which could have been the cause for slower RTs especially for the small set size and distant proximity. Consequently, one of the problems in creating large and small set-sizes is that crowding increases density and density increases with set-size. When keeping the density constant, mean eccentricity increases with set-size, which is an unwanted by-product (Cohen & Ivry, 2001). As a result, an increase in mean eccentricity should have created longer RTs and as it has been noted; bigger set-sizes should take more time due to bigger saccades i.e. eye movements (Monnier P, Nagy L. A, 2001). Subsequently, Meinecke & Donk (2002) have suggested that one should be careful when using different variations in set-sizes in visual display searches as analysis could be based on comparisons of two unequal and non-interchangeable conditions, which could be a possible factor in no main significant effects.
To conclude, although set-size seems to be confounded with other parameters such as density and spatial regularity (due to proximity), further research on this topic is necessary. Consequently, even although the results were not as initially predicted, this is a fruitful step in the right direction of looking at the different processes behind proximity and distracters in visual search tasks. There are methodological problems that have been mentioned that can be improved upon in future research with the optimistic possibility of finding more desired results.
Nothdurft H.-C. (1999). Focal attention in visual search, Vision Research, 39 (14), pp. 2305-2310.
Nothdurft H.-C. (2002). Attention shifts to salient targets, Vision Research, 42 (10) , pp. 1287-1306.
Pöder, E. (2007). Effect of colour pop-out on the recognition of letters in crowding conditions. Psychological Research/Psychologische Forschung, 71(6), 641-645. doi:10.1007/s00426-006-0053-7
Treisman, A., & Sato, S. (1990). Conjunction search revisited. Journal Of Experimental Psychology: Human Perception And Performance, 16(3), 459-478. doi:10.1037/0096-1518.104.22.1689
Treisman, A. M., & Gelade, G. (2000). A feature-integration theory of attention. In S. Yantis (Ed.) , Visual perception: Essential readings (pp. 347-358). New York, NY US: Psychology Press.
Schubö A, Schröger E, Meinecke C (2004) Texture segmentation and visual search for pop-out targets: An ERP study, Cognitive Brain Research, Copyright © 2004, Elsevier, http://dx.doi.org/10.1016/j.cogbrainres.2004.06.007
Monnier P, Nagy L. A (2001) Set-size and chromatic uncertainty in an accuracy visual search task; Vision Research, Copyright © 2001, Elsevier, http://dx.doi.org/10.1016/S0042-6989(01)00240-1
Whitney D, Levi M. D (2011) Visual crowding: a fundamental limit on conscious perception and object recognition, Trends in Cognitive Sciences, Copyright © 2011, Elsevier, http://dx.doi.org/10.1016/j.tics.2011.02.005
Dakin C. S, Cass J, Greenwood A. J, Bex, J P (2010) Probabilistic, positional averaging predicts object-level crowding effects with letter-like stimuli Journal of Vision August 17, 2010 vol. 10 no. 10 article 14, doi: 10.1167/10.10.14
Meinecke C, Donk M, (2002), “Detection performance in pop-out tasks: Nonmonotonic changes with display size and eccentricity” Perception 31(5) 591 – 602
Wolfe, J. M., Oliva, A., Butcher, S. J., & Arsenio, H. C. (2002). An Unbinding Problem? The disintegration of visible, previously attended objects does not attract attention. Journal of Vision, 2(3): 256-271