Effects of flanker type and position on foveal letter recognition

Participants

Eight participants were involved in this study, based on convenience sampling. All participants had best-corrected visual acuity of 6/6 or better in both eyes, healthy, and had no significant ocular or systemic diseases. This research was approved by Universiti Kebangsaan Malaysia’s Researc Ethics Committee (NN-2014-083), which complied with the tenets of the Declaration of Helsinki. All participants provided written informed consent before the start of the experiment.

Apparatus

The experiment was run on a Macintosh desktop, where Matlab (version R2012a) with PsychToolbox version 3.0.8 extension (Pelli, 1997) was installed to create the experimental stimuli. Stimuli were digitised Sheridan Gardiner letters (A, H, O, T, U, V, and X), scanned from their printed clinical chart format. Each letter had a 5×5 construction, that is, each stroke was one fifth of the letter size and each letter had equal width and height. The Weber’s contrast of the letters was 90%. Stimuli were displayed on a Samsung CRT monitor running at 100 Hz.

Procedure

In this study, we measured contour interaction and crowding functions for visual acuity (Flom et al., 1963). First, we measured performance of each participant to identify an isolated target letter. The physical size of the letter displayed on the computer screen was fixed. Isolated letters were presented randomly for 100 trials and participant’s responses were recorded. A viewing distance was chosen (between 11 and 13 metres) so that the percent correct response fell between 80 to 90 percent for each participant. All participants completed four 100-trial runs. Performance across the four runs was averaged to obtain the unflanked (i.e. uncrowded) performance.

Next, the participant’s task was to identify the target letter in the presence of crowding features. When the target letter was flanked by bars, the separation between them was expressed in edge-to-edge separations of 0, 1, 2.5, 3.5, 5, and 10 stroke-widths (each stroke-width is equivalent to one fifth of the target letter size). The bars were placed either horizontally, that is, to the left and right of the target letter (Figure 1a); vertically, that is, above and below the target letter (Figure 1b); or at the top, bottom, left and right to the target letter (Figure 1c).

Figure 1.

a. Target letter H flanked horizontally by two vertical bars. The size of each stroke of the letter (x) is one fifth of the whole letter size (which is 5x). The width of the flanking bars is one stroke-width and the length of the bars is the same to that of the target letter. The separation shown here is one stroke-width, where separation is measured from the closest edge of the target letter to the closest edge of the flanking bars. b. Target letter T flanked vertically by two horizontal bars. c. Target letter A flanked by four bars on all its sides. d. Target letter V flanked horizontally by two other letters. The centre-to-centre separation between target and flankers shown here is one letter width, which is equivalent to edge-to-edge separation of zero stroke-width, i.e. the target abuts the flankers. e. Target letter X flanked vertically by two other letters. f. Target letter U flanked on four sides by four other letters. g. Target letter O enclosed in a crowding box. The stroke-width of the box is the same as the stroke-width of the target letter. The separation between the closest edge of the letter and the box shown here is one stroke-width.

For letter flankers, the separation between the target letter and its flankers were calculated as the distance from the centre of the target to the centre of one of the flanking letters. Hence, separations between them were expressed in centre-to-centre units, which were 1, 1.25, 1.5, 1.75, 2 and 3. Note that centre-to-centre separation of 1 is equivalent to zero stroke-width in edge-to-edge units (i.e., two closest edges of the target and flanking letters were abutting). The flanking letters were placed either horizontal to the target letter (to the left and right of the target letter, Figure 1d); vertical to the target letter (above and below it, Figure 1e); or, at the top, bottom, left and right of the target letter (Figure 1f). Additionally, performance was also measured when the target letter was enclosed in a box (Figure 1g). The stroke of the box was one-fifth of the target letter size. The crowding features were always of the same contrast as the target letter. When letters were used as the crowding feature, they were never the same letter as the target and were always different from each other.

Table 1 summarises the separation between target and flankers used in the study, expressed in edge-to-edge units for bar flankers and surrounding box, and centre-to-centre units for letter flankers. Conversion from centre-to-centre units to edge-to-edge units is also shown, to allow direct comparisons of the effects of the different types of flankers on letter recognition.

All participants completed four runs of each crowding condition. There were 100 trials in each run. Within each run, the target letter and the separation between the target and its crowding feature was randomised. In this part of the experiment, isolated letters (i.e. unflanked) were also included in some of the trials, with an equal probability of being displayed as the flanked letter trials. All experimental runs were conducted in a systematic order. Runs with flanking bars were completed first, followed by runs with flanking letters and runs with the surrounding box. In total, each participant completed 3200 trials, excluding practice trials.

Participants indicated their response verbally. The response was then entered by the examiner using a keyboard. No feedback was given. Viewing time was unlimited, that is, participants were allowed to take their time to identify the target letter, although generally the responses were prompt. With each keyboard input, the stimulus display was replaced by a mean luminance screen that appeared for 500ms before the presentation of the next trial. Testing was done monocularly using the subject’s dominant eye, determined with the hole-in-hand test (Miles, 1930). The non-viewing eye was covered with a black occluder. All participants were given training with the task and data collection only commenced after they were comfortable with the experimental procedures (at least two hours of training). Data collected during the training period were not included in the final analysis.

Statistics

Raw data were sorted in MS Excel version 14 and plotted in IgorPro^TM version 6.3.6.4 (WaveMetrics, Portland, OR). Statistical analyses were carried out using IBM Statistical Package for Social Sciences (SPSS) version 19.0. Data were analysed using repeated measures Analysis of Variance (ANOVA) with Greenhouse-Geisser correction to compare change in performance obtained with the different flanker types and positions. Post hoc pairwise comparisons with Benferroni corrections were performed where appropriate. Alpha level of 0.05 was used for all statistical tests.

Results

Figure 2 shows relative performance to identify the target letter as a function of target-flanking bar separation. Relative performance was calculated by subtracting the performance with flanking bars from performance to recognise an unflanked target. Therefore, performance lower than 0% correct for a flanked condition indicates a performance lower than that found for unflanked condition, i.e. crowding. Likewise, performance greater than 0% correct for a flanked condition indicates that participants’ correct responses for a flanked condition were higher than an unflanked condition.

Figure 2. Relative performances (% correct response) to identify the target letter flanked by two bars placed to the right and left of the target (i.e. horizontal arrangement, open diamonds), and above and below of the target letter (i.e. vertical arrangement, open triangles).

Relative performance was also measured for a target flanked by four bars (open circles). Performances are shown as a function of target-flanker separation.

Repeated measures ANOVA revealed significant main effects of position [F_{(1.67, 11.71)} = 9.43, p = 0.005] and target-flanker separation [F_{(1.74, 12.17)} = 21.05, p < 0.001) on relative performance. The interaction between the position of flanking bars and target-bar separation was significant [F_{(3.58, 25.07]} = 3.95, p = 0.02]. That is, the change in relative performance for identifying the target across separation was influenced by the position (and number) of the flanking bars. Two flanking bars placed at the bottom and top of the target (open triangles in Figure 2) exerted the least crowding effect that was significantly different from that obtained with four flanking bars [F_(1,7) = 13.07, p < 0.01] . Performances with two flanking bars placed to the left and right of the target (open diamonds in Figure 2) and four bars surrounding the target (open circles in Figure 2) were not significantly different [F_(1,7) = 5.14, p = 0.06]. The extent of crowding was also different dependent on the position of the flankers. Post hoc tests revealed that performance with top and bottom bars was significantly lower from unflanked conditions at separations of 0 (p=0.005) and 1 (p=0.001). With left and right bars, performance was significantly lower from unflanked conditions at separations of 0 (p<0.001), 1 (p=0.003) and 2.5 (p=0.006) stroke widths. With four surrounding bars, performance was also significantly lower from unflanked conditions at separations of 0 (p<0.001), 1 (p<0.001) and 2.5 (p=0.004) stroke widths.

Figure 3 shows relative performance to identify the target letter as a function of separation between the target and flanking letters. Repeated measures ANOVA revealed that there was significant main effects of flanker position [F_{(1.36, 9.54)} = 10.53, p = 0.006] and separation F_{(1.98, 13.88)} = 62.54, p < 0.001] on relative performance. There was a significant interaction between flanking letter position and target-flanker separation [F_{(2.79, 19.56)} = 6.54, p = 0.004]. Unlike with flanking bars, letter flankers placed at the top and bottom of the target letter (open triangles, Figure 3) exerted as much crowding as, and not significantly different from, that obtained with four letter flankers (filled circles in Figure 3) placed around the target [F_(1,7) = 0.02, p = 0.88]. Least reduction of performance was obtained by two letter flankers that are placed to the left and right of the target (open diamonds in Figure 3) and it was significantly different from that obtained with four letter flankers [F_(1,7) = 47.41, p < 0.001]. Post hoc tests showed that performance with left and right letter flankers was significantly lower than unflanked performance at separations of 0 and 1.25 centre-to-centre. For top and bottom, and four surrounding letter flankers, performance was significantly lower from unflanked conditions at separations of 0, 1.25 and 2.5 centre-to-centre (all p < 0.05). That is, the extent of crowding was larger for these two stimulus configurations.

Figure 3. Relative performances (% correct response) to identify the target letter flanked by two other optotypes placed to the right and left (open diamonds), above and below (open triangles), of the target letter, and four optotypes surrounding it (filled circles) as a function of target-flanker separation.

Figure 4 compares relative performances in identifying the target when it was flanked by four bars, four flanking letters and when it was surrounded by a box. For all types of flankers, the largest reduction in performance occurred when the flankers abutted the target. Generally, participants’ relative performance to identify the target letter was similar for a target flanked by either four letters or four bars. At separation of one stroke width, the surrounding box appeared to exert the least crowding effect compared to other types of flankers. Repeated measures ANOVA reveals that the interaction between flanker type and separation is significant [F_{(3.42, 23.92)} = 3.15, p = 0.038), that is, the change in relative performance across separation was dependent on the flanking features used. Post hoc analyses revealed that a significant difference in performance with the different flanker types only occurred between letter flankers and box, when they abutted the target letter (p = 0.007), but not for bar flankers and at all other separations. The difference between flanked and unflanked performances occurred at a separation of 0, 1 and 2.5 stroke widths. This extent of crowding was similar to that found when four bars and four letter flankers were used.

Figure 4. Relative performances (in percent of correct response) to identify a target letter flanked by four bars (open circles), four optotypes (closed circles) and a surrounding box (open squares), as a function of target-flanker separation (in multiples of stroke width).

Discussion

Our results are consistent with earlier findings that flanking bars reduced the recognition of a foveally fixated acuity target when they are in close proximity to each other (e.g. Flom et al., 1963; Stuart & Burian, 1962). Regardless of the type of the flankers used recognition of a target letter was similarly affected when there were four flankers surrounding it. Our results obtained with four surrounding letters and bars are consistent with the observations of Norgett & Siderov (2014) in adult observers. Other studies have also reported that with adult foveal viewing, crowding effects were similar for flankers that are categorically similar and dissimilar to the target (Danilova & Bondarko, 2007; Leat et al., 1999; Song et al., 2014). Norgett & Siderov (2014) reported that in children however, at separation half a letter width away (i.e. 2.5 stroke-widths) significantly larger foveal crowding effects were observed with letter flankers compared to four flanking bars. In our study, other than in the abutting condition, the largest reduction of performance occurred with target-flanker separation of one stroke-width. It is likely that larger crowding effects in children than adults would also be observed with flankers placed at one stroke-width away from the target.

We also showed that the reduction in correct performance was dependent on the number and relative position of the flankers surrounding the target. Other than the abutting condition, separation of about one stroke-width, edge-to-edge, exerts the most crowding effect for all conditions. However, at this target-flanker separation, the position of the flankers relative to the target letter influenced the magnitude of the crowding effect. Liu (2001) reported different patterns of performance for a square C orientation discrimination task when it was flanked by two bars that was dependent on the flanker’s position relative to the gap of the C, i.e. either parallel or orthogonal to the C’s gap. Therefore, if a crowded chart is to be developed using bars as the crowding feature, maximum crowding can be exerted with flanking bars placed either to the left and right or on all four sides of the target letter with a separation of one stroke width.

With two letter flankers, reduction in performance is significantly dependent on their position relative to the target letter. We observed the least crowding when the letters were place to the left and right of the target letter, especially when they were abutting and at approximately one stroke-width away from the target (Figure 3), compared to when they were placed above and below the target letter. Asymmetric horizontal-vertical crowding region have been reported when target and distractors are presented in the peripheral visual field (Feng et al., 2007; Hairol et al., 2014; Toet & Levi, 1992). An interesting observation is that two letter flankers placed at the top and bottom of the target letter reduced the participants’ performance just as much as having four letter flankers surrounding the target. Cherici et al. (2012) reported that for normal adults, prolonged fixation was inaccurate during foveal fixation, and the eye movement range when one was fixating may be larger than the spacing between letters at size threshold. It could be assumed that crowding would be similar regardless of whether a target letter is flanked vertically or horizontally by other letters. Indeed, a similar amount of crowding has been reported for a single letter surrounded by four letters on all its sides and for a string of five letters surrounded by other letters (Norgett & Siderov, 2014). Bedell et al. (2015) reported that more errors were made by adult observers when identifying a long string of letters compared to a short string of letters (all were arranged horizontally). However, it is unclear how error rate would differ for a string of vertically arranged letters. We showed that performance was reduced further for a target letter flanked vertically by other letters than that for letters that are flanked horizontally (Figure 3). As most adults are presumably more trained for reading letters that are arranged horizontally (as was the case for all participants in this study) there could be less fixation inaccuracies when letters are read horizontally compared to when the letters are read vertically. It has been shown that reading rate is similar for experienced adults who are trained to read either horizontal or vertical texts (Sun et al., 1985; see also Rayner, 1998). This implies that horizontal-vertical asymmetry in crowding is not necessarily a fixed characteristic but most likely it is experience- and skill-dependent.

Particular attention should be given to the type of flankers used to induce crowding and contour interaction effects. We showed that when a surrounding box was placed at 2.5 stroke-widths (half a letter width) away from the target, the magnitude of crowding was not significantly different from that produced by either four letters or four bars, indicating that at this particular target-flanker separation, similar outcome would be found regardless of the flanker type. However, when placed at one stroke-width away from the target letter, flanking letters and bars caused approximately a further 10% reduction in performance to identify the target letter, compared to a surrounding box although this difference in performance did not reach statistical significance. Similar results have also been reported previously (Lalor et al., 2012). This may be of clinical implication since a number of crowded letter charts, designed to induce crowding, use a box as a crowding feature to surround the target letter (e.g. McGraw & Winn, 1993; Salt et al., 2007).

Various acuity tests designed to induce crowding typically used bars or other letters as their crowding feature, such as the logMAR crowded test, the Cambridge Crowding Cards and Sonksen LogMAR test. The separation between the target letter and the crowding feature typically is half an optotype width (but for the Sonsken logMAR test, the separation is one letter width). Formankiewicz & Waugh (2013) suggested that crowding in acuity tests could be enhanced by placing flankers closer to the target letter than in the currently available test. Our results complement this suggestion, where we also showed that larger magnitude of crowding can be obtained, at least for recognising single letters, if the crowding feature is placed at a distance closer than those designed in commercially available charts, specifically at one stroke-width away (0.2 letter width) from the target. At this particular separation, similar crowding magnitude was obtained with either letter or bar flankers. Indeed, larger foveal crowding effects have been reported for flankers that are placed closer than half a letter width away from an acuity target (Flom, 1991; Formankiewicz & Waugh, 2013; Lalor et al., 2016; Norgett & Siderov, 2014). Therefore, this separation could be considered in future designs of clinical acuity charts, so that these charts could be more sensitive in detecting eye conditions that suffer from crowding such as amblyopia. However, a surrounding box at this same separation produced approximately 10% less reduction of performance compared to other types of flankers. It has been reported that crowding reduces when flankers are grouped together, possibly aiding target recognition (for example, extending bars to form a surrounding box) (Banks & White, 1984; Manassi et al., 2012; Sayim et al., 2011).

In this study, abutting condition consistently produced the maximum amount of crowding, regardless of the flanker number or type. However, one must be careful when considering it in the design of clinical acuity charts. It has been observed that performance to localise the gap of a Landolt C was better when flankers abutted the target, compared to when the flankers were about 1 or 2 stroke widths away (Danilova & Bondarko, 2007; Flom et al., 1963). However, it was also reported that abutting flankers elicited the largest crowding magnitude compared to other target-flanker separations for localising the gap of a square C (Hairol et al., 2013) and for acuity measured with picture and letter optotypes (Formankiewicz & Waugh, 2013). Abutting may be a special case of crowding. When flankers and the target are just abutting, this combination may change the overall appearance of the target letter. It has been shown that dioptric blur reduces the amount of crowding obtained in abutting condition (Formankiewicz & Waugh, 2013) which would predict that those with uncorrected refractive error (or perhaps anisometropic amblyopia) would perform better than those with normal vision. Hence realistically, the spacing that exerts crowding is preferably not when the flankers are touching target.

Conclusions

Recognition of letters in foveal viewing is influenced by the distance, type and number of flankers surrounding them. When a letter is surrounded by only two letter flankers, crowding magnitude differ significantly, depending whether they are in vertically- or horizontally-arranged configurations. This effect is possibly due to imprecise eye movements when extracting the target from the flankers, which could be experience- and skill-dependent. The magnitude of crowding is similar for a target surrounded by flankers at all four sides, regardless of the flanker type, at a target-flanker separation of half optotype width, which is widely used in commercially available crowded clinical charts. Maximum crowding can be induced at target-flanker separation of one stroke width (equivalent to 0.2 optotype width) when the target is surrounded on all four sides by either bars or other letters. A box however produces slightly less crowding effect, although not statistically significant, at this particular target-flanker separation. Our results have implications on the future designs of acuity tests that aim to exert the maximum crowding effect, so that they become more sensitive in detecting certain eye conditions such as amblyopia.

Data availability

F1000Research: Dataset 1. Percent correct performance for target identification, 10.5256/f1000research.8572.d120563 (Hairol et al., 2016).

Consent

Written informed consent for publication of their clinical data was obtained from the participants.