An investigation into the effects of crowding

Published: Last Edited:

This essay has been submitted by a student. This is not an example of the work written by our professional essay writers.

Abstract: Crowding occurs when objects are positioned so close to each other that it becomes impossible to separate their individual features. The relationship between the spacing of the flanking letters which induce crowding and the point of fixation is quantified by Bouma's law. The mechanisms behind the phenomenon are not thoroughly understood, though it has been speculated that it occurs during higher-order visual processing. An investigation was conducted to discover whether the positioning of flankers had any effect of the magnitude of crowding. Flankers were positioned either above-and-below (vertically) or side-by-side (horizontally) the target letter. Three theories were considered, either a difference in results between horizontal and vertical; radial and tangential; or another unknown but definite pattern, called X-theory. A definite pattern was found when 16 observers were tested in four peripheral visual locations with a series of stimuli with differently positioned flankers, either horizontally or vertically about the target letter. The observers had to identify the target letter amongst the flanking letters, and nearly all found this task much harder when the flanking letters were arranged radially (radiating from the fixation point) about the target. The results were found to be highly significant and were virtually uniform across the observers. There was a marked difference in the vertical meridian, whilst horizontally, the results for both orientations of flankers were more similar, hinting at a possible connection with our ability to read text in this direction. [235 words]

Fig. 1. A physical demonstration of crowding. By fixating on the letter 'X', it is much easier to identify the individual letter 'R', since it is relatively close and is uncrowded. The task is much harder in the lower example despite the position of the 'R' remaining the same. The task must, therefore, have increased in difficulty due to the presence of the flankers (letters 'C' and 'H').

Introduction: The crowding phenomenon occurs when an observer struggles to discriminate small objects or stimuli when they are in close proximity to one another. The phenomenon occurs as a result of 'contour interaction' (Millodot, 2009) and its existence has been proven by the observers ability to easily identify the same stimuli when no crowding 'flankers' are present. This can be reproduced by viewing Fig. 1, and noting how much easier the task of recognising the letter 'R' when it is uncrowded is compared to when crowding letters, or 'flankers' surround it. When applied to vision in general, crowding is most often experienced by a patient when difficulties with reading arise, as the close arrangement of small letters induce the contour interaction. The experience perceived is that of a blending of the stimuli, so that the observer is perceptually aware of their presence, but is unable to separate the individual characters. Crowding remains consistent with different objects of regard, and the phenomenon occurs not only with optotypes, but also with facial recognition, Vernier acuity (the ability to report whether two lines are orientated in a level plane) and also moving targets (Levi, 2008). It has been discovered by Andriessen & Bouma in 1976, that crowding does not affect the actual detection of a target (Levi, 2008).

Fig. 3. The possible crowding theories. The central spot in each diagram represents the fixation target and the lines indicate the position of the flankers. a) and b) illustrate the Vertical-Horizontal theory respectively, c) and d) show the Radial-Tangential theory whilst e) represents X-theory, where the optimum flanker orientation is unknown.

Fig. 2. Flankers and critical spacing. Bouma's Law stipulates that the maximum critical spacing is equal to roughly half the eccentricity (in this case the distance between the fixation 'X' and letter 'R'). The letters 'C' and 'H' are known as flankers and it is the presence of these which induce the effect referred to as 'crowding', since they make identification of the letter 'R' a much harder task.

Crowding, despite arising through a relatively simple process, remains mostly mysterious. Little is known for definite about exactly how the process occurs, though much of the modern literature (Levi, 2008; Pelli & Tillman, 2008) focuses on post-retinal processing being the possible key to the solution. This can be either in the neurone receptive fields, where the distracting flankers fill the inhibitory surround of a receptive field, thus detracting from the signal caused by the letter being identified falling on the excitatory central region (Tripathy & Cavanagh, 2002); or within the visual cortex (mostly beyond the primary visual cortex or V1 area) where stimuli features are independently detected and combined. The information from the stimuli is registered in the visual cortex, and despite a great ability in recognising letters, because the cortex is unable to separate the features into the individual objects, an incomprehensible 'jumble' is visualised instead, thus the problem could possibly lie not in feature detection (as stated above, crowding is not reported to have an effect on detection), but in feature combination and therefore the subsequent identification process cannot take place.

The effect of crowding, however, is not cut-and-dry. The 'optimum' spacing of flankers, i.e. the distance where they produce the crowding effect, varies with retinal eccentricity. This is quantified by the Bouma Law (Pelli & Tillman, 2008), where the optimum distance (or maximum spacing between the target letter and the crowding letters), known as the 'critical spacing', is equal to roughly half of the target eccentricity and the increase towards peripheral vision arises due to the increased spatial pooling in the outer reaches of the retina. Any position closer to the object will produce the crowding effect. The flankers' crowding ability is also dependent upon their similarity to the target stimuli trying to be identified. If they have a different contrast, then the crowding effect is greatly reduced to almost non-existent in some circumstances. If they are vastly different in feature and appearance, then the effect of crowding is lessened also. In the central region of foveal vision, crowding requires such a small critical spacing that it is often not reported at all (Levi, 2008).

What has not yet been truly investigated is whether the positioning of the flankers at a given location has any overall effect on their 'crowding effectivity'; in other words does it matter whether the flanking letters are side by side the target letter, or above and below it?. By studying this, it is hoped that a general theory will be applicable to flanker orientation and how this affects vision. There are several possible theories which may or may not be correct, and these will be referred to accordingly, and demonstrated above in Figure 3; Horizontal-Vertical theory, whereby no matter what position parafoveally the stimuli are presented (Up, Down, Left, Right), there is a preference towards either flankers oriented horizontally or vertically, meaning that one of these has a much more marked crowding effect; Radial-Tangential theory, where the effectivity of the flankers depends on whether they are positioned radially (along a plane radiating from the central fixation point) or tangentially (positioned perpendicular to a plane radiating from the fixation point); or X-theory, where there is a definite pattern to the results, but it fits into neither of the previous theories. In the example of Figs 1 and 2, the flankers are positioned radially and horizontally. It is expected that there will be a definite pattern to the results, and that this will fall into one of the above theories, thus meaning the remaining two can be eliminated.

Methods: Observers: 16 observers were used in the study, 10 females and 6 males, all aged between 19 and 26. All 16 observers were optometry students and were familiar with the experiment and what was being investigated. All observers wore their habitual refractive correction and all but one had functioning binocular vision (one participant had gross suppression in the right eye).

Equipment: Two laptop PCs were used, both running on the Windows XP platform and the experiment was run using a specially designed presentation on Microsoft Office 2007 PowerPoint. Due to the difference in screen sizes, one laptop was chosen as the 'standard' to which all others used would be calibrated to. This was achieved through use of a calibration slide whereby a square of known and fixed dimensions was viewed as a slideshow and measured from the working distance already established. A working distance of 57cm was chosen as it was both practical and allowed for a direct correlation between size and visual angle (1cm = 1° of visual angle). The size of the calibration square on the second laptop was used to calculate the adjusted working distance, which ensured that overall the optotypes presented did not alter in size.

Stimuli: The stimuli were to be presented at four parafoveal locations surrounding a centre fixation spot. These were 'Up', 'Down', 'Left' and 'Right' and the target letter (i.e. the one to be identified by the subject) was always positioned with a 7° eccentricity from the centre of fixation. This avoided any problems of either the target letter or flankers falling on the blind spot and also kept the test within the confines of the display upon which the stimuli were presented. The stimuli were presented in a high contrast environment, with black optotypes appearing on a uniform white background. The stimuli dimensions were kept constant at 8.3cm in height and width, which translated to a visual angle of 5° in both directions, and the font used was Arial Bold, since this is a universal font and most closely resembles that used in modern logMAR visual acuity charts. The stimuli were presented for a three second period, with a gap between stimuli of a further three seconds.

A PowerPoint slideshow was created with individual slides for each presentation. For each of the four parafoveal locations, nine presentations of stimuli with differently orientated flankers were shown. Two control slides were also created, where only the target letter appeared without flankers present. Therefore, for each location there were a total of twenty slides (nine with horizontal flankers, nine with vertical flankers and two control slides), making the final PowerPoint presentation consist of eighty examination slides. The positioning of the stimuli and flankers is summarised in Fig 4. These were then randomised and their order noted for results gathering.

A pilot study was run in order to find constant values for other parameters within the experiment, primarily the critical spacing of the flanking letters. This was taken as a distance of 1.25cm between the centres of the target letter and the flanking letter, since a score of 50% or more was recorded for all observers and so eliminated the possibility of correct identification occurring through chance. The pilot study also highlighted that each observer had to give a response, since to be unsure of the letter produced a confounding variable of observer confidence in letter identification. During the final test it was stressed that the observer must give a response for each presentation.

Fig. 4. Stimuli positioning and flankers. 'X' denotes the central fixation point, and 'T' the target letter (i.e. that being identified by the observer). The 'V's at each location represent the vertical flanking letters, whilst the 'H's represent the horizontal flankers. Furthermore, green letters denote the flankers positioned radially, whilst red letters represent the tangential flankers.

Procedure: The observer was seated in the laboratory with good illumination and a comfortable sitting posture. The laptop was positioned at the correct working distance and the screen tilted to ensure it lay perpendicular to the primary visual axis when the observer was viewing the fixation target. The purpose of the experiment was explained and a short demonstration presentation was shown in order to fully inform the observer and ensure that the task was completely understood. Any questions or concerns that the observer had were addressed before continuing the experiment. The results were taken by a fellow experimenter, seated to the side of the observer. This allowed the recorder to keep track of which stimulus was being presented as during the pilot study this had proved problematic. An additional experimenter was positioned opposite the observer to monitor fixation, and make notes of the number of fixation losses encountered during the experiment. The experiment was run twice for each observer, each sitting at the same computer as before.

Results: Having repeated the experiment twice for each participant, a collection of 32 samples of data was obtained. When tabulated, this gave overall percentages for each subject's performance at each of the four parafoveal locations. The results showed that the performance of all but one subject (incidentally, this was a different subject for each location) greatly followed one of the outlined trends in the original hypothesis. The control stimuli for each direction all gained high percentage rates of being correctly identified, so strengthening the reliability of the rest of the results obtained. A t-test was conducted on the results in order to show whether the data collected was significant (Armstrong & Hilton, 2011). The results are summarised in Figure 5, the statistical analysis shown in Figure 6, and the average results displayed graphically in Figure 7. For each of the tested locations, the results were found as such:

Fig. 5. Graph illustrating the average results for each meridian including standard error bars for each bar. The data is based on the percentage of correct responses given by the 16 observers for each direction, thus the lowest bar in each direction illustrates the flanker orientation to give optimum crowding effectivity. The diagrams at the base of each bar provide reference to the positioning of the flanking letters relative to the letter being identified.Up: Here, the results greatly favoured the horizontal flankers, meaning that a greater proportion of target letters were correctly identified with flankers positioned in this direction. The results vary considerably, as demonstrated by the high (22.87 and 28.45 for horizontal and vertical respectively) standard deviations which are much lower in the other directions. The standard errors for this orientation are also relatively high when compared to the other results. With a two tailed t-test performed on the data, the p value was found to be so small that it has been represented simply as "<0.001" adding further strength to these results, since they are highly significant findings.Graph 2.jpg

Down: The difference in results is larger than for 'Up', though here the control average was brought down slightly. Again, preference for horizontal (or tangential) flankers is clearly evident as the percentage correct is over 90% compared to below 60% for the opposite meridian. The standard deviation is lower here, as are the standard errors. With the t-test performed on the 'Down' results the data is again found to be statistically significant (p=<0.001).

Left: The strongest individual preference was found with the 'Left' data, as the average score was 96.87% with flankers positioned vertically. There is also a smaller gap between the horizontal and vertical results for this data, and the standard errors and deviations are much lower than the 'Up' and 'Down' results. The results were found to be statistically significant when a t-test was run (p=<0.001).

Fig. 6. Tabulated average percentages for correct identification of the target letter in the different directions tested. Each figure is displayed to two decimal places. The preferred orientation in each direction is stated beneath, meaning that this orientation offered the weakest crowding effect. Below are the theoretical preferences for each of the initial hypotheses. Right: The data recorded for stimuli presented on the right of the observer again showed a strong preference for vertical flankers, although there is the smallest difference between flanker orientation at this location. The standard errors and deviations are slightly higher than those for the 'Left' data, though they are still low values. The t-test revealed that the data was highly significant, though it was higher than all of the other meridians (p=0.0013).

Fig. 7. Tabulated statistical analyses of each direction's data. Note the p values for each meridian are incredibly low, highlighting the high significance of all of the data recorded. The relevance of the results collected is elaborated upon further in the 'Discussion' section below.

Discussion: The results from the data obtained showed a very strong preference that was identical across the majority of the observers. The preference of the flanker orientation was shown to favour that of tangential flankers since for all four directions, the flankers positioned perpendicular to those points (i.e. horizontal flankers in the vertical meridians, vertical flankers in the horizontal meridians) elicited the weakest crowding effect meaning radial flankers elicit the strongest crowding effect. This is reflected in the highest results in each meridian (see Fig. 6) being for the orientations relating to tangentially placed flankers. These results were all found to be highly significant, since each p value was at least 0.001, illustrating that the likelihood of these results (and subsequently the t-test values) occurring through chance, considering the null hypothesis to be correct, was reduced to one tenth of a percent. For all directions examined, the null hypothesis could be rejected with almost definite certainty, proving that the orientation of flanking letters does affect the crowding experience. Further interpretation of the results showed that it was tangential flankers with the highest rate of correct identification, meaning that the Radial-Tangential theory was therefore the most significant. This means that when flankers are positioned radially, no matter what direction, they cause the greatest crowding effect and thus make object recognition hardest. It also means that both the Horizontal-Vertical theory and X-theory can both be eliminated from the analysis since the preference shown by the observers does not favour either of these theories. Examination of both the results (Fig. 6) and also the graph (Fig. 5) show that the greatest difference in flanker effectivity occurred in the vertical meridian (Up and Down locations). Along the horizontal meridian, although a definite trend is present, it is much less pronounced. This may have arisen as a result of how the observers have learnt to read, in that they read sentences which have been formed of letters arranged horizontally. What this does not explain, unfortunately, is why the crowding effect in the horizontal meridian, is at its greatest when the flankers are positioned in a setting which is most like reading. This may account for the theories of how reading is performed by the visual system, that the eye makes saccadic jumping movements and constructs what is read from only a small series of letters at the beginning and end of the word (Schad et al. 2010).

Since the null hypothesis has been rejected with a great level of certainty, the phenomenon must now be investigated further in order to attempt to uncover the mechanisms which cause this Radial-Tangential crowding discrepancy, and why radial flankers specifically cause such a problem in object recognition. A number of subsequent studies could be carried out to further understand this phenomenon and perhaps shed more light onto the complex visual processing which occurs within the visual cortex. This is an important insight that should be pursued in order to help understand vision in more depth. One most significant finding may come from the comparison of performance between western and oriental ethnicities, since those examined in this study were all of a western background, where reading is a task which requires letters to be positioned in a horizontal arrangement, whilst those from the Far East tend to read characters which are arranged vertically. Though one would expect this phenomenon to have given preference to the Horizontal-Vertical theory of crowding, it may still prove to be significant and is worth further investigation, especially since performance for all observers was much greater in the horizontal peripheral zones (i.e. Left and Right (see Fig. 5)) than the vertical, both inter-orientation (horizontal and vertical flankers) and inter-direction (left-right performance was higher than that of up-down). A possible finding may be that the phenomenon is opposite for those who are accustomed to reading vertically, and it may be tangential flankers which affect vision the most. Another study could involve the use of tracer materials to illustrate the particular areas of the visual cortex that are being stimulated during the task, with the eventual use of microelectrodes to take single-cell recordings of particular areas in order to discover the exact location of the confusion caused by crowding and what produces the flanker orientation preference. In short, whilst this particular study has shown with a great level of certainty that flanker orientation does indeed have an effect on how well we are able to recognise a target, it has only taken the world of visual science a step closer towards complete understanding, and it will involve many more investigations before it will be possible to say that the visual system is completely understood.

Whilst the experiment was conducted in a controlled environment with as much consistency as possible maintained, there were still several flaws to the experiment which may possibly need to be addressed in a follow-up study if there is any doubt to the results shown in the above tables. Firstly, and possibly most noteworthy, is the sample of observers. Despite being very strong results obtained with the relatively small sample, (this would change very little if the number of observers were increased since the likelihood of these results occurring through chance have already been proved to be remote) the sample should be expanded to include subjects of varying ethnicities and ages since both of these were relatively uniform in the experiment. Particular attention should be paid to those who are of an Eastern origin, as mentioned before, in order to discover whether this phenomenon is related at all to the way in which we read. Secondly, further experimental modifications may need to be carried out to reduce the fixation errors given by the subject, or at least instigate a form of objective measurement. During the experiment conducted, the fixation was assessed through the subjective monitoring of an experimenter. A possible source of this problem was the saccadic movements of the eyes, which tend to arise when a stimulus is suddenly presented in the periphery which causes an involuntary movement of the eyes, since the role of peripheral vision is to alert the visual system to the presence of an object and subsequently foveate onto it for closer analysis. Subjective observation meant that these movements, despite lasting only several milliseconds and so possibly not being long enough to gain any actual information from the stimuli, may be misidentified as the subject losing fixation and actively looking for the stimulus. The method of an inbuilt illumination system which monitors the position of a corneal reflex on the subject as used in visual field analysis would lend itself well to this experiment to give a much more consistent and objective result for fixation. Thirdly, and least significant, but worth noting for completion, the slide arrangement on the PowerPoint presentation which was shown to subjects was not entirely random, since an experimenter had to manually mix the slides; meaning that theoretically, they could have learnt the arrangement and answers to the stimuli. This is, however, unlikely and could be easily avoided by not letting any of the experimenters take part in the study.

Overall, despite the flaws within the study, the data collected was shown to be highly significant and that the Radial-Tangential theory was correct, since overall the strongest affecter of crowding was radially positioned flankers. As mentioned above, subsequent studies can now be conducted with the focus being aimed at flankers positioned in these two orientations to further shed light on how the visual system processes grouped objects. Whilst the effect of crowding may simply be a curiosity left over from our evolutionary antecedents or possibly a small quirk of the human visual system it remains to be seen, but further studies should help the field of vision science to further understand the complexities of our own vision.