The effect of motion on crowding: Zooming text

Abstract

Crowding is the major sensory factor responsible for the slow reading speeds exhibited in peripheral vision. Past attempts to improve peripheral reading via crowding reduction have generally focused on applying spatial changes to the stimulus and have been largely ineffective. Recent evidence indicates that dynamic approaches have good potential for reducing crowding in peripheral reading. We tested this hypothesis by introducing “zooming” motion (smooth letter resizing across the presentation duration) to trigram stimuli (groups of three randomly selected letters) presented at 10° in the lower visual field and evaluating recognition of the middle letter. Crowding was alleviated in the presence of this motion, both when dynamic cues were introduced to all letters in the trigram simultaneously and when they were applied to individual letters alone. The magnitude and direction of crowding reduction depended on the amplitude and direction of motion. These results suggest that dynamic presentation may be a useful tool for improving peripheral reading through reducing letter crowding. Zooming motion, in particular, has the additional advantage of conserving text layout, making it a good candidate for such an application.

Introduction

Crowding, a fundamental limiting factor of reading and object recognition in peripheral vision (Pelli et al., 2007; Whitney & Levi, 2011; Yu, Legge, Wagoner, & Chung, 2014), refers to a phenomenon in which identification accuracy for target objects is reduced when presented in close proximity to other objects (see Figure 1 for an example demonstration). 

Figure 1

 

Demonstration of crowding, a phenomenon in which object identification is deteriorated when target objects are surrounded by nearby flanking objects. When fixating the red dot on the left, the middle letter of the trigram (string of three random letters) below is hard to decipher. By comparison, fixating the red dot on the right and identifying the isolated letter below is easy. In our experimental context, stimuli were always presented 10° below fixation.

A key feature of crowding is its dependence on eccentricity (Bouma, 1970; Pelli, 2008). Crowding is minimal at the fovea, but becomes increasingly pronounced as objects are presented further into the periphery (Bouma, 1970; Flom, Weymouth, & Kahneman, 1963). Although the exact mechanism underlying crowding is unclear, accounts of crowding generally involve the incorrect integration of features across objects. This incorrect integration of features may be a result of pooling or averaging of information (features or their positions) across a receptive field (Balas, Nakano, & Rosenholtz, 2009; Dakin, Cass, Greenwood, & Bex, 2010; Freeman, Chakravarthi, & Pelli, 2012; Greenwood, Bex, & Dakin, 2009; Parkes, Lund, Angelucci, Solomon, & Morgan, 2001), or may be the result of positional uncertainty or spatially unfocussed attention that results in the inappropriate migration or substitution of features from flanking to target objects (Chastain, 1982; Huckauf & Heller, 2002; Wolford, 1975). 

Since crowding poses a substantial challenge to peripheral vision, various ways have been explored to mitigate crowding. One simple way of reducing crowding is to increase the distance between target and flanking objects. Bouma (1970) suggested that as long as the spacing between objects is increased with eccentricity to maintain a proportion of roughly one half of the eccentricity or more, crowding remains minimal. Crowding also appears contingent on the spatial similarity between the target and the flanking objects (see Whitney & Levi, 2011 for a review). Crowding can be lessened through the introductions of target-flanker differences along a wide range of spatial properties such as color (Kooi, Toet, Tripathy, & Levi, 1994; Põder, 2007), contrast polarity (Chung & Mansfield, 2009; Kooi et al., 1994), spatial frequency (Chung, Levi, & Legge, 2001), and size (Nazir, 1992). In the temporal domain, crowding was found to be mitigated when the target object differed from flankers in temporal properties such as onset time (i.e., generating temporal spacing) (Huckauf & Heller, 2004; Scolari, Kohnen, Barton, & Awh, 2007), temporal frequency (Bex & Dakin, 2005), and the presence or absence of transient blink (Greenwood, Sayim, & Cavanagh, 2014). 

Crowding is the key process responsible for the slow reading speeds exhibited in the periphery (Pelli et al., 2007; Yu et al., 2014). Past attempts to improve peripheral reading via crowding reduction have generally focused on introducing spatial changes to the stimulus, such as increasing letter spacing (Chung, 2002), enlarging the spacing between lines of text (Bernard, Anne-Catherine, & Eric, 2007; Calabrèse et al., 2010; Chung, 2004; Chung, Jarvis, Woo, Hanson, & Jose, 2008), and decreasing the spatial similarity of adjacent letters (Chung & Mansfield, 2009). However, despite the critical importance of spacing and spatial similarity to crowding, these manipulations have limited effectiveness in improving peripheral reading. In the combined spatial-temporal domain, recent evidence lends support to the use of displacement motion (where different retinal areas are stimulated over the presentation duration) in assisting with peripheral reading. A small amount of retinal-image jitter has been shown to improve letter recognition accuracy in the periphery for normally-sighted people under both isolated and crowded conditions (Yu, 2012), and to improve peripheral word recognition amongst patients with central vision loss (Watson et al., 2012). 

Why might the addition of motion in the stimulus be beneficial? Peripheral vision has been found to be more sensitive to stimuli containing dynamic changes than to stationary stimuli (Foster, Gravano, & Tomoszek, 1989; Sharpe, 1974; To, Regan, Wood, & Mollon, 2011). There is evidence showing that appropriate amounts of retinal motion improve peripheral visual acuity (Brown, 1972; Macedo, Crossland, & Rubin, 2008). Additionally, transient dynamic changes can interfere with image fading, a substantial problem (especially in peripheral vision) during rigid fixation (Martinez-Conde, Macknik, Troncoso, & Dyar, 2006; Rucci, Iovin, Poletti, & Santini, 2007), which may in turn enhance stimulus detection in contexts that might otherwise induce fading. Another possible beneficial factor is that motion may help promote precise spatial coding of letter features and correct feature binding through coherent motion trajectory and/or potentially enhanced local attention, which may consequently reduce crowding and improve reading. 

The present study explores the influence of motion on crowding. Beyond any advantages that dynamic information may provide in peripheral tasks, motion also offers another distinct design advantage over spatial changes—leaving the spatial properties of the stimulus largely unchanged. For example, introducing retinal-image jitter to a letter has little impact on the color, shape, or spatial-frequency of the letter. Motion allows us to maintain the spatial properties of presented letters at optimal settings for reading (such as presenting letters in high contrast), and yet still introduce dissimilarities between letters that might facilitate de-crowding of the stimulus.  

Motion can be introduced to a stimulus in a variety of ways. Here we focus specifically on a “zooming” manipulation, where letters are smoothly resized while maintaining their positions stably across the presentation duration. As Figure 2 shows, the zooming motion gives rise to a percept similar to using the zoom function in a computer document. Like the displacement motion introduced in previous studies (Watson et al., 2012; Yu, 2012), zooming motion similarly introduces changes in retinal stimulation over time. It comprises expansion/contraction movements which are orthogonal to displacement motion. This manipulation was chosen because it confers an advantage over the other dynamic manipulations studied in preserving text layout. Text layout (arrangement of letters) provides useful information such as word shape (Tinker, 1963) and by extension, sentence form. Ideally, this information should be maintained with the introduced dynamic changes. The zooming manipulation can achieve this by introducing dissimilarity across letters without disrupting the text layout. As explored in Experiment 1, letter positions and spacings were fixed throughout the letter resizing. 

Figure 2

 

Example stimulus manipulations in Experiments 1, 2, and 3. Each zooming manipulation was achieved by a five-frame animation. The first column depicts an example zoom manipulation where center-to-center spacing was maintained across the five-frame stimulus presentation (Experiment 1). The middle two columns demonstrate zooming of the target letter and of the flanking letters respectively (Experiment 2). The last column depicts example zoom manipulations where center-to-center spacing was scaled proportionately with letter resizing (Experiment 3). This Figure depicts only examples of the 0.5° zoom range. In the case of forward zooming, letter sizes start small (0.5° in this Figure) and increase across the five stimulus frames with the final size as 1°, whereas, for reverse zooming, letter sizes start at 1° and reduce in size across frames.

In this study, we conducted a series of experiments to assess the effect of the proposed manipulation on crowding. We began by establishing in Experiment 1 that zooming motion (referred to simply as “motion” in the remainder of the text) can reduce crowding and enhance the identification of the target letter when motion is applied to all letters simultaneously (full zooming). Experiment 2 then examined whether the effect size observed in Experiment 1 can be increased by selectively adding motion to portions of the stimulus (partial zooming). Although Experiment 2 introduces additional within-frame dissimilarity cues (size differences between letters at a given moment in time), comparisons between Experiments 1 and 2 permit an examination of the relative contributions of spatial (size difference) and temporal (motion) factors to crowding reduction. Experiment 3 then considers whether motion can be implemented in the “simplest” way possible by resizing the entire stimulus proportionally (proportional zooming). Proportional zooming is visually equivalent to moving a book or page of text closer and further away from oneself. 

Methods

Participants

The three experiments were conducted with separate groups of subjects: Six subjects (mean age 22.7 years, range 19-29 years) participated in Experiment 1, six subjects (mean age 20.2 years, range 19-21 years) participated in Experiment 2, and five subjects (mean age 24.4 years, range 19-33 years) participated in Experiment 3. All subjects were English speakers with normal or corrected-to-normal vision, and were naive to the experimental stimuli and purpose (with the exception of one subject in Experiment 3, who was an author of the study). The study was conducted in accordance with the Declaration of Helsinki and was approved by the OSU Institutional Review Board. Subjects gave written informed consent prior to the experiment. 

Apparatus and stimuli

Stimuli were presented on a ViewSonic Graphics Series G225f CRT monitor in a dark room. A viewing distance of 40 cm was maintained by use of a chin rest. The effective screen size was set to be 30.1 × 22.5 cm for Experiment 1 and 39.9 × 29.4 cm for Experiments 2 and 3. The resolution was 1024 × 768 pixels with a frame rate of 90 Hz throughout all experiments. Experiments were produced and displayed using Matlab R2010a and Psychtoolbox 3 (Brainard, 1997; Kleiner et al., 2007; Pelli, 1997). A presentation duration of 222 ms was used in all experiments, with an additional duration (444 ms) tested in Experiment 1 only. 

The stimuli were trigrams (strings of three random lowercase English letters) constructed using Courier font with a largest print size of 1°. Print size is defined as the height of the lowercase letter x. The trigrams were always presented at an eccentricity of 10° below fixation in −99% Weber contrast (black type against a white background). The trigrams were presented at six logarithmically-spaced letter spacings (1.0°, 1.2°, 1.7°, 2.2°, 2.9°, and 3.9°) within each experimental condition. The chosen letter spacings covered a range that included standard letter spacing (a center-to-center separation of 1.16 × x-width for Courier font) and estimated spatial extent of crowding (also known as critical spacing). Spatial extent of crowding has been reported to be about 40% to 50% of the target eccentricity for radially positioned flankers (Bouma, 1970; Whitney & Levi, 2011; Yu et al., 2014). Since the ratio of radial to tangential spatial extent is approximately 2.5:1 along the vertical meridian (Toet & Levi, 1992), we expect that the spatial extent of crowding for the static condition in the tangential direction is around 1.6° to 2° at 10° eccentricity below fixation. Our results did fall within this range (1.7° for the static condition, averaged across all three experiments). 

Experiment 1: Full zooming

The primary manipulation in this experiment was the introduction of motion to the entire trigram stimulus (see Figure 2 for an example stimulus manipulation). To zoom the letters, the letter size was smoothly altered across the presentation duration in either a forward or a reverse direction. Forward zooming letters began with a smaller print size and increased in size with a constant speed across the presentation until they achieved full print size (1°) at the final stimulus frame, whereas reverse zooming letters began at full (1°) print size and decreased in size across the presentation. Center-to-center letter spacings were fixed throughout the letter resizing. A static control condition was also included, where all letters maintained a print size of 1° throughout the presentation. 

The range of the zoom was manipulated by varying the smallest presented print size (the starting size in the forward zoom, and the terminal size in the reverse zoom). The lower size limits employed were 0.875°, 0.75°, 0.5°, and 0.25°. With an upper size limit of 1°, these lower limits result in zoom ranges of 0.125° (1°–0.875°), 0.25°, 0.5°, and 0.75° respectively. Image resizing always occurred across five stimulus frames with fixed duration (e.g., 44.4 ms/frame for a total presentation duration of 222 ms) regardless of the zoom range. As a result, larger zoom ranges appeared to change size more quickly (zoom velocity = zoom range/presentation duration). Table 1 lists print sizes used in the five stimulus frames for each zoom range. 

Table 1

 

Print sizes used in the five stimulus frames for each zoom range.

Zoom ranges	Size 1	Size 2	Size 3	Size 4	Size 5
0.125°	0.875°	0.906°	0.938°	0.969°	1°
0.25°	0.75°	0.813°	0.875°	0.938°	1°
0.5°	0.5°	0.625°	0.75°	0.875°	1°
0.75°	0.25°	0.438°	0.625°	0.813°	1°

Subjects completed three experimental sessions of 18 blocks. Each session was split into two halves on the basis of presentation duration (222 ms or 444 ms). The order of the two half-sessions was alternated across sessions and was counterbalanced across subjects. Within each half-session, the order of the nine blocks, corresponding to nine experimental conditions, was completely randomized. The nine conditions comprised two zoom directions (forward or reverse) by four zoom ranges (0.125°, 0.25°, 0.5°, and 0.75°) and one static condition. Within each block, the six letter spacings were randomized across 90 trials. Since only 222 ms was tested in Experiments 2 and 3, the current analysis focused primarily on the data obtained at 222 ms, with only brief discussion of the results from the longer duration (444 ms). 

On each trial, the subject pressed a mouse button to trigger the stimulus (trigram) presentation while maintaining stable fixation at the fixation point provided. The task was to identify the middle letter of the trigram. No feedback was provided to the subject about their letter identification accuracy. During testing, eye movements of subjects were monitored for compliance by the experimenter who could reliably detect 2° saccades. Whenever eye movements were detected, trials were cancelled and reshuffled into the trial sequence. On average, less than 1% of the trials were discarded. 

Experiment 2: Partial zooming

In Experiment 2, we selectively introduced motion either to the target letter alone, or to the flanking letters alone (Figure 2). These two conditions (target versus flanker zooming) were carried out in two separate half-sessions, counterbalanced across sessions and subjects. Otherwise, the stimulus and procedure were identical. 

Experiment 3: Proportional zooming

In Experiment 3, one critical change was made to the stimuli: As trigram letters were re-sized over time, center-to-center spacing was allowed to vary proportionally (see Figure 2). This alteration was accomplished by resizing the entire trigram as a unit (uniform resizing), rather than resizing each letter individually. This manipulation involves a combined motion of zooming (all three letters) and translation (flankers only). Because the entire trigram was scaled as a unit within any given stimulus frame, all three letters were the same size, similar to Experiment 1. This manipulation is advantageous because it provides the simplest mode of application in real-world scenarios. For example, moving a page of text toward and away from oneself would induce this kind of proportional resizing of the letters and their spacing. We employed the same four zoom ranges used in Experiments 1 and 2. In addition, we included a wide range of static conditions: 0.25°, 0.5°, 0.625°, 0.75°, 0.875°, 0.94°, and 1° letter sizes. These static letter sizes were chosen to include the smallest and largest letter sizes and the average letter size of each zoom range (e.g., 0.94° is the average letter size for the zoom range of 0.125°), permitting direct comparison of performance between zooming stimuli and static stimuli of comparable sizes. There were two sessions in this experiment. Subjects completed a total of eleven blocks (seven static conditions and four zoom conditions) in each session, with 15 trials per letter spacing in each block. 

Data analysis

Our primary measure of interest was crowding reduction—the extent to which motion reduced crowding relative to the static condition (see Figure 3). This measure was determined as follows. First, the spatial extent of crowding was determined for each condition. Specifically, percent correct was plotted against letter spacing, and 75% thresholds were obtained by fitting Weibull functions using the Palamedes Toolbox (Prins, 2012; Prins & Kingdom, 2009). Thresholds were then normalized to the static condition¹ (zoom range = 0°). Crowding reduction was calculated using the equation  

 

for each zoom range (0.125°, 0.25°, 0.5°, 0.75°). Crowding reduction was then plotted against zoom range and fitted with a parabolic function. The parabolic function had two free parameters and the constraint that the parabola must pass through the coordinate (0, 0). This constraint was imposed because the static condition was defined as having a zoom range of 0° and was expected to have no change in crowding (i.e., 0% crowding reduction), with crowding in all other conditions defined relative to this fixed static condition. Based on the parabolic fit, the magnitude of the peak reduction of crowding and the corresponding zoom range could be estimated (see results for Experiment 1). Figure 3 illustrates the fitting process with example data from one subject for the target-zoom condition. Figure 3a shows the fitting of the raw data and the estimations of thresholds. Figure 3b shows the plotting of these thresholds across zoom range. Figure 3c shows the rescaling of these threshold values relative to the static condition to produce the crowding reduction measure. Positive values indicate reduction of crowding relative to the static condition.  

Figure 3

 

Example of data processing. (a) Letter-recognition accuracy versus letter spacing for the static condition (red) and the four zoom conditions. The spatial extent of crowding is indicated by the 75% threshold. (b) Spatial extent of crowding (75% threshold) as a function of zoom range. (c) Crowding reduction (the proportional reduction in crowding relative to the static condition) as a function of zoom range. Positive values indicate reduced crowding.

Results

Experiment 1: Full Zooming

Experiment 1 investigates whether crowding can be reduced by introducing motion simultaneously to all letters in a crowded trigram stimulus. Figure 4 depicts crowding reduction as a function of zoom range (results from 222 ms duration only are shown). A repeated-measures ANOVA was conducted with factors: two directions (forward vs. reverse) × four ranges (0.125°, 0.25°, 0.5°, 0.75°). The results of this analysis indicated that, overall, crowding reduction was significantly greater for forward than for reverse zoom, F(1, 5) = 45.20, p < 0.001. Crowding reduction was also significantly affected by zoom range, F(3, 15) = 10.25, p = 0.02. The interaction between zoom direction and range was not significant. For forward zooming, crowding was reduced when a moderate motion (zoom range = 0.25°) was applied to the stimulus, t(5) = 5.24, p = 0.003, one-sample t test. The smallest zoom range did not differ significantly from static. At larger zoom ranges, performance either did not improve or became marginally impaired relative to static. No crowding reduction advantages were observed for reverse zoom. 

Figure 4

 

Results of Experiment 1 (presentation duration = 222 ms). Crowding reduction as a function of zoom range for individual subjects (top two rows) and group average (bottom row) for forward (red) and reverse (blue) zoom conditions. Error bars represent ± 1 standard error of the mean.

To more closely examine the effect of zoom range on performance, we analyzed the vertex location of each parabolic fit. The x coordinate of the vertex indicates the zoom range at which the peak occurs (i.e., the zoom range that produces the maximum crowding reduction), while the y coordinate of the vertex indicates the magnitude of the crowding reduction. Of interest was whether crowding was reduced relative to the static condition. Because we constrained the parabolic fit to pass through coordinate (0, 0), this reduces to determining whether the x and y coordinates of the vertices were different from zero for each zoom direction.² When zooming was introduced in a forward direction (where the letter sizes increased over time), the vertices were mostly located away from the static condition along the x axis. Maximum crowding reduction occurred at a zoom range of 0.22° ± 0.05° (SE). This zoom range was significantly different from static, t(5) = 4.23, p = 0.008, and was associated with a crowding reduction of 7.5 ± 2.5% which differed significantly from zero, t(5) = 3.02, p = 0.03. By contrast, reverse zooming was ineffective at reducing crowding (peak crowding reduction = 2.1 ± 1.1%). Even when subjects were permitted a longer presentation duration of 444 ms, these effects persisted (i.e., a significant crowding reduction at the vertex for forward but not reverse zoom conditions). Maximum crowding reduction occurred at 0.26° ± 0.07° zoom range, t(5) = 3.77, p = 0.01 with a magnitude of 6.3 ± 2.2%, t(5) = 2.88, p = 0.04. A paired samples t test showed that the zoom range that produced the maximum crowding reduction did not differ between the two presentation durations, indicating that crowding reduction may be selectively tuned to zoom range and associated spatial-temporal characteristics other than zoom velocity (i.e., zoom range/presentation duration). 

Experiment 2: Partial zooming

Since the effect size in Experiment 1 was relatively modest, Experiment 2 aimed to determine whether the selective introduction of dynamic motion to the target or flanking letters can introduce a greater magnitude of crowding reduction. Crowding reduction was analyzed using a three-way, repeated-measures ANOVA (two zoom locations (target vs. flanker) × two zoom directions (forward vs. reverse) × four zoom ranges). As is apparent from Figure 5, the effect was notably different when motion was added to the target letter (solid lines) versus when it was added to the flanking letters (dashed lines). These differences were reflected in significant interactions between zoom location and zoom direction, F(1, 5) = 19.73, p = 0.007, and between zoom location and zoom range, F(3, 15) = 9.33, p = 0.004. No other interactions were significant. The two significant interactions, together, indicate that the introduction of motion to the flanking letters successfully reduced crowding, and the effect on crowding increased with zoom range, regardless of the direction of that motion. However, when motion was introduced to the target letter, crowding was reduced for forward zoom but increased for reverse zoom at moderate to large zoom ranges. 

Figure 5

 

Results of Experiment 2. Crowding reduction as a function of zoom range for group averaged data. Data are presented for two zoom directions: forward (red circles) and reverse (blue diamonds), and two zoom locations: zooming of target alone (filled symbols) versus flankers alone (open symbols). Parabolic functions were used to fit the data. Error bars represent ± 1 standard error of the mean.

Comparison of Experiments 1 and 2

To compare the results of Experiments 1 and 2, Figure 6 overlays the results of forward zoom conditions from both experiments. Interestingly, the magnitude of crowding reduction at a zoom range of 0.25° was very similar across all three forward zoom conditions. The whole-trigram zoom did not differ from the target-only zoom, t(9) = 0.072, p = 0.94 nor from the flanker-only zoom, t(9) = 0.048, p = 0.65. In the case of the larger zoom ranges, crowding was consistently reduced relative to static when forward-zooming motion was introduced to the target or flankers, but worsened (increased) when all three trigram letters were zoomed in the same manner. The maximum magnitude of crowding reduction within the measured range of zooming was 7.5% for whole-trigram zoom, 23.7% for target-only zoom, and 40.6% for flanker-only zoom, averaged across subjects. The magnitudes for partial zooming were comparable to those obtained in the crowding studies manipulating target-flanker similarity in the spatial domain (e.g., Kooi et al., 1994), whereas the magnitude for full zooming was considerably smaller. Together, these results suggest that the larger crowding benefits (reductions) observed in Experiment 2 (particularly at large zoom ranges) were due to the combined effects of temporally varying letter size (motion), and of letter-size differences within each stimulus frame of the trigram. 

Figure 6

 

Mean crowding reduction for forward zoom conditions from Experiment 1 (full zooming) and Experiment 2 (partial zooming) overlaid.

Experiment 3: Proportional zooming

The goal of Experiment 3 was to determine whether the crowding-reduction benefits of motion extend to the case where the resizing occurs proportionally across letter size and letter spacing. Figure 7a presents crowding reduction as a function of zoom range in Experiment 3. The data appear to follow a similar trend to that observed in Experiment 1 (with a small reduction in crowding for short zooming manipulations). However, a one-way, repeated-measures ANOVA of the effect of zoom range on crowding indicated that this trend was only marginal if not absent, F(4, 16) = 3.402, p = 0.09. 

Figure 7

 

Results of Experiment 3. (a) Crowding reduction as a function of zoom range, relative to the 1° static condition. The data were fit with a parabolic function. (b) Crowding reduction as calculated relative to static conditions that were size-matched to the mean letter size (annotated inside data bars) for each zoom range. In both plots, positive values indicate reduced crowding relative to the static condition. Error bars represent ± 1 standard error of the mean.

The inclusion of multiple static conditions in this experiment provided us with several baselines for evaluating crowding reduction. In particular, by including static conditions size-matched to the average letter size across the zoom, we could ask whether the motion added anything to the subjects' performance above and beyond what would be expected based on the average letter size alone. Figure 7b replots the same data as presented in Figure 7a, but scaled relative to the average letter size. The size of the static reference for each condition is annotated inside the data bars. Crowding reduction in this context was defined as  

 

for each zoom range (0.125°, 0.25°, 0.5°, 0.75°). If the effect of presenting a moving stimulus were similar to the effect of viewing the same average letter-size across the presentation, the crowding reduction measure would be zero. Instead, in all cases, the measure was significantly greater than zero: Performance in the zoom condition outperformed that expected by the average letter size alone in every case. These results imply that motion provides a performance benefit compared to the mean-size-matched static condition, even when the spacing is sized proportionally. Nonetheless, the effect size as measured through the static size of 1° was smaller than that observed in Experiments 1 and 2.  

Discussion

We found that crowding could be reduced in the presence of zooming motion both when motion was introduced to the entire trigram simultaneously (Experiment 1) and to a greater degree when motion was introduced partially at the level of individual letters (Experiment 2). This reduction was not symmetrical across zooming directions. When motion was applied to the target letter (irrespective of the conditions of the flankers), crowding was alleviated for forward zoom but not for reverse zoom. This result was true at least for moderate zoom ranges. The crowding reduction was minimal when center-to-center spacing was varied proportionally with letter size during zooming (Experiment 3). 

Mechanisms of crowding reduction underlying zooming

We found that the magnitude of the crowding reduction across experiments was highly dependent on zoom range. When the zoom range was moderate, the crowding reduction caused by forward zooming of the entire trigram (Experiment 1) was similar in magnitude to that observed when forward zooming was applied to subcomponents (letters) of the trigram (Experiment 2). However for large zoom ranges, the only effective method of reducing crowding was to introduce motion to selective subcomponents of the trigram. These results suggest the possibility of two independent mechanisms of crowding reduction: temporal cues (e.g., effects directly related to the motion itself), and spatial cues (e.g., effects related to size-differences on individual stimulus frames). We will consider each factor in turn. 

How might temporal cues induce the observed crowding reduction? One possibility is that the motion trajectory permits better tagging of object features to their individual letters, akin to the Gestalt principle of grouping by “common fate” (Wertheimer, 1923). Common or coherent motion as a mode of binding or grouping isolated elements to construct global forms or objects has been demonstrated across a wide range of phenomena including contour integration (e.g., Ledgeway & Hess, 2002), 2D form from motion (e.g., Regan, 1986), 3D structure from motion (e.g., Treue, Husain, & Andersen, 1991), and biological-motion point-walkers (Johansson, 1973). In the current study, the simultaneous resizing of the trigram letters induces motion trajectories, with each letter in the trigram having its own trajectory from its own point of origin. The separate motion trajectories associated with each letter could serve to appropriately segregate and tag the letter features, reducing the feature integration errors inherent to crowding. In a study examining the effect of motion trajectories on target identification in random-dot kinematograms, grouping by common fate was found to be weaker for convergent than for divergent motion (Stürzel & Spillmann, 2004). Consistent with this report, we observed that forward zooming (which results in divergent motion) is more effective in inducing crowding reduction than is reverse zooming (which results in convergent motion). 

Another possibility is that the motion induces changes in deployment of spatial attention. Attention has been demonstrated to modulate the effect of crowding (Chakravarthi & Cavanagh, 2009; Dakin, Bex, Cass, & Watt, 2009; Freeman & Pelli, 2007; Strasburger, 2005; Yeshurun & Rashal, 2010). Attention can be captured by stimulus motion to a peripheral locus (Abrams & Christ, 2003; Franconeri & Simons, 2003, 2005; von Mühlenen & Lleras, 2007). More specifically, the dynamic signals of expanding stimuli (but not contracting stimuli) have been shown to guide spatial attention and facilitate performance in visual search (Franconeri & Simons, 2003), detection (Takeuchi, 1997; von Mühlenen & Lleras, 2007), and letter discrimination tasks (von Mühlenen & Lleras, 2007) in the periphery. Ecologically, expanding stimuli (such as approaching objects) are more likely to require an immediate reaction than contracting stimuli (receding objects) (Franconeri & Simons, 2003). Therefore, forward zooming may have more attentional priority than reverse zooming even with matched stimulus frames or motion magnitude. The bias towards expansion has also received support from neural studies (Ptito, Kupers, Faubert, & Gjedde, 2001; Tanaka & Saito, 1989; Wunderlich et al., 2002). If deploying spatial attention to the peripheral target location ameliorates the effect of crowding, we expect forward zooming (expanding) to be more effective in inducing crowding reduction than reverse zooming (contracting). Our observations appear to be consistent with this prediction. 

The second factor that may have contributed to crowding reduction in this study is within-frame spatial cues. This factor was relevant only when motion was introduced to selective portions of the trigram (Experiment 2), inducing size-based dissimilarity cues within any stimulus frame. Size differences alone between targets and flankers appear to induce crowding reduction (Nazir, 1992). In the present study, the common magnitude of crowding reduction across Experiments 1 and 2 for small zoom ranges (e.g., 0.25°) suggests that, at these ranges, the shared motion trajectory and/or enhanced local attention predominantly dictated the degree of crowding reduction. However, at larger zoom ranges (e.g., 0.5°), the effect of relative sizing within frames predominated. In Experiment 1, where the size cue was absent, no crowding reduction was observed for large zoom ranges. In Experiment 2 where the size cue was available, crowding reduction was observed and was greater in magnitude than for smaller zoom ranges, suggesting that the size-based spatial cues had greater salience than the temporal cues. 

Possible neural mechanisms

The neural mechanisms underlying crowding are not yet well understood. As the psychophysical research indicates, crowding can take place at multiple stages in the visual hierarchy (see review by Whitney & Levi, 2011). Neuroimaging findings are consistent with this view and indicate that possible loci of crowding in form perception (e.g., alphabet letters) may reside between V1 and V4 (Anderson, Dakin, Schwarzkopf, Rees, & Greenwood, 2012; Bi, Cai, Zhou, & Fang, 2009; Fang & He, 2008; Freeman, Donner, & Heeger, 2011; Millin, Arman, Chung, & Tjan, 2013). 

In the current study, we have established that the introduction of motion has the effect of reducing letter crowding under certain conditions. What might this imply about mechanisms of crowding? Form and motion perception are widely believed to be primarily supported by two distinct neural architectures (e.g., Goodale & Milner, 1992; Merigan & Maunsell, 1993; Mishkin, Ungerleider, & Macko, 1983): the ventral stream (predominantly responsible for form-based tasks) and the dorsal stream (predominantly responsible for motion or action based tasks). Recently, a growing body of evidence suggests that the dichotomy of processing between form and motion is not as clear as initially assumed and that interaction between the two streams is probable (see review by Cloutman, 2013). Dorsal stream signals are thought to be conveyed more rapidly than ventral stream signals (Chen et al., 2007; Nowak, Munk, Girard, & Bullier, 1995; Schroeder, Mehta, & Givre, 1998), leading to the proposal that faster dorsal processing could support feedback information to the ventral stream (e.g., Bullier, 2001). Such a mechanism could potentially underlie the type of motion-supported form-processing that we have demonstrated in the current study. 

Proportional zooming

Of the tested procedures, the proportional zooming is the most practical given that movement of a fixed block of text (e.g., moving a book toward and away from oneself) is the simplest way to implement such motion. Compared to full zooming (Experiment 1), proportional zooming differs with respect to the relationship between size and spacing. The smaller crowding reduction observed with this procedure is consistent with the existing knowledge on the importance of letter spacing to crowding: within the spatial extent of crowding, the amplitude of crowding is inversely dependent on the spacing between the target and its flanker(s). In Experiment 3, we also found that motion reduces crowding relative to static stimuli of average letter size, but not relative to static stimuli at the largest letter size (1°). These results indicate that to reduce crowding in the context of the proportional zoom procedure, the motion needs to straddle the target letter size. Given the notable benefit found at the average size of 0.94° (zoom range was between 0.875° and 1°), we expect a similar benefit to exist for letters with an average print size of 1°. 

Effect of letter size

In the forward-zoom condition of Experiment 1, moderate zoom ranges facilitated crowding reduction, but large zoom ranges impaired performance. Since the zoom manipulation employed a wide range of letter sizes across image frames, it is reasonable to ask whether performance was particularly impaired by large zoom ranges simply because of the smaller average letter size. Previous studies suggested that, in the periphery, spatial extent of crowding does not depend on target size as long as it is above recognition level (Levi, Hariharan, & Klein, 2002; Pelli, Palomares, & Majaj, 2004; Tripathy & Cavanagh, 2002). To evaluate the effect of letter size, we conducted two additional control experiments to determine (a) the threshold size needed to recognize individually presented static letters and (b) whether performance accuracy for isolated letter recognition is affected by the introduction of motion. The first of these controls aimed to determine whether the individual letters in the presented trigrams were visible (suprathreshold) across the presentation duration. If the letter sizes for some proportion of the presentation duration are below acuity threshold, the spatial extent of crowding may be dependent on target size, which could be a factor in explaining the poorer performance associated with large zoom ranges. This control also indirectly evaluated the relative difficulty of individual letter identification across the different zoom manipulations. The second of these controls was designed to directly determine whether the particular motion applied in these experiments can impair individual letter performance by testing individual letter recognition under the same conditions applied to the experimental trigrams. Figure 8 presents the results of these two control experiments. As shown in Figure 8a, the size threshold (visual acuity) for recognizing statically presented single letters was 0.62° ± 0.025°. For large zoom ranges such as 0.5° and 0.75°, one or two out of the five stimulus frames were likely below subjects' acuity threshold. In other words, the stimulus might not have been discriminable during the entire stimulus presentation for large zoom range conditions. The second control experiment directly addressed the impact of this difference in individual letter discriminability. As shown in Figure 8b, the ability to discriminate isolated letters was largely unaffected by zoom range. Performance accuracy for single letter identification remained high (>90%) throughout, suggesting that the introduction of motion did not notably impair letter recognition. This result indicates that despite the sub-threshold letter sizes included in large zoom ranges, the effect of motion on individual letter identification is insufficient to explain the observed effects on crowding variation. 

Figure 8

 

Results from two control experiments with 11 subjects (combined subjects from Experiments 2 and 3). (a) Static letter identification accuracy for letter sizes 0.2°, 0.25°, 0.35°, 0.5°, 0.75°, 0.875°, and 1° (30 trials per letter size). Individual data are plotted in grey, and group averages with standard error bars are in black. Mean threshold (75% criterion) was 0.62° (SE = 0.025°). (b) Zooming letter identification accuracy for 1° letters presented at four zoom ranges (0.125°, 0.25°, 0.5°, and 0.75°) in forward (black solid lines) and reverse (grey dashed lines) directions. Performance is close to ceiling across conditions, with no significant effect of range nor zoom direction.

Across the three experiments reported here, all zooming stimuli were presented at sizes close to or smaller than the critical print size (CPS), the smallest print corresponding to maximum reading speed (e.g., Chung, Mansfield, & Legge, 1998). It is possible that the benefit from motion would be greater if the entire zooming range were presented at sizes that were “suprathreshold” relative to the CPS. 

Clinical implications

Since crowding is predominantly a problem of peripheral vision, it has a particular impact on clinical populations with central vision loss. Patients with central vision loss must rely on peripheral vision to perform tasks such as reading. Although enlarging peripherally presented text can assist with problems of reduced acuity (Chung et al., 1998; Latfiam & Whitaker, 1996), text enlargement does not address the additional constraints imposed by crowding. Peripheral reading remains slower than foveal reading even with larger text (Chung et al., 1998). Since crowding is the major sensory factor limiting reading (Yu et al., 2014), successful methods for reducing crowding of peripheral stimuli have potential implications for improving peripheral reading. Overall, attempts to improve peripheral reading through crowding reduction have met with limited success. The dynamic methods explored here would be difficult to immediately apply to a clinical population because the effect size of crowding reduction was relatively small under the tested conditions. These results, rather, offer a proof in principle that dynamic variations in stimuli can reduce crowding. Future work is needed to address whether larger, more clinically meaningful improvements in crowded letter recognition can be induced by variations on the tasks presented here, and whether the resulting crowding reduction is beneficial to peripheral reading. 

Summary and conclusions

Our investigation established that the addition of dynamic cues such as zooming motion can alleviate the effect of crowding on letter recognition in peripheral vision. The benefit can be observed when moderate dynamic cues are applied to one or more presented letters with conserved center-to-center spacing, conditional on the direction of zooming. The crowding reduction is possibly due to better tagging of features to their individual letters, assisted by motion trajectory cues and/or potentially enhanced local attention. Given the advantage of conserving text layout, this type of motion seems a good potential option for improving reading performance in the periphery through crowding reduction. 

Acknowledgments

This study was supported by Seed Funding of Interdisciplinary Projects from the Center for Cognitive and Brain Sciences at OSU. We are grateful to Sheng He, Changbin Huang, and Zhong-Lin Lu for their helpful discussions.  

Commercial relationships: none. 

Corresponding author: Deyue Yu. 

Email: [e-mail]. 

Address: College of Optometry, Ohio State University, Columbus, Ohio, USA. 

Footnotes

¹  In Experiment 2 where we split the experimental session into two separate “half-sessions” to separately assess zooming for flankers and for targets, we measured static performance twice, once in each half-session. Even though these two static blocks were identical, we chose to measure flanker-zooming and target-zooming benefits relative to their own separate static blocks for a better control of local testing context.

Footnotes

²  In cases where the vertex was beyond the range of the data, the x-axis coordinate corresponding to the maximum y value was taken to be the maximum point of crowding reduction.