Adaptation to a spatial offset occurs independently of the flash-drag effect

Abstract

Abstract

Abstract:

Abstract  Visual motion can influence the perceived position of an object. For example, in the flash-drag effect, the position of a stationary flashed object at one location appears to shift in the direction of motion presented at another location in the visual field (Whitney & Cavanagh, 2000). The results of previous physiological studies suggest interactions between motion and position information in very early retinotopic areas. However, it is unclear whether the position information that has been distorted by motion further influences the visual processing stage at which adaptable position mechanisms may exist. To examine this, we presented two Gabor patches, each of which was adjacent to oppositely moving inducers, and investigated whether adaptation to the illusory spatial offset caused by the flash-drag effect induced the position aftereffect. Our results show that a change in the perceived offset in the presence of the flash-drag effect did not influence the position aftereffect. These results indicate that internal representations of positions altered by the presence of nearby motion signals do not feed into the mechanism underlying the position aftereffect.

Introduction

Visual motion can influence the perceived position of an object. In one example, hereafter called the motion-induced position shift (MIPS), the apparent position of the blurry contour of a region that contains moving dots or a drifting sinusoidal luminance modulation is perceived to shift in the direction of the motion (De Valois & De Valois, 1991; Ramachandran & Anstis, 1990). In another example, hereafter called the flash-drag effect (FDE), the position of a stationary flashed object at one location appears to be shifted in the direction of motion presented at another location in the visual field (Whitney & Cavanagh, 2000). These illusions have been extensively studied to clarify how visual motion interacts with position perception. Regarding the MIPS, it has been shown that not only luminance-defined motion, but also contrast-defined motion (Bressler & Whitney, 2006), color-defined (isoluminant) motion (Ashida, Yamagishi, & Anderson, 2007), cyclopean motion (Murakami & Kashiwabara, 2009), and plaid motion (Hisakata & Murakami, 2009; Mather & Pavan, 2009) can induce the illusory position shift. Regarding the FDE, substantial evidence suggests the operation of higher-order motion mechanisms, including global motion (Scarfe & Johnston, 2010), inferential motion (Watanabe, Nijhawan, & Shimojo, 2002; Watanabe, Sato, & Shimojo, 2003; Whitney, 2006), and attentive tracking (Shim & Cavanagh, 2004, 2005; Tse, Whitney, Anstis, & Cavanagh, 2011). On the other hand, Whitney (2005) demonstrated that adaptation to motion without awareness of the motion's direction yields a shift in the perceived position of a stationary stimulus presented at the same location in the absence of awareness of any motion aftereffect. Fukiage, Whitney, and Murakami (2011) used randomly phase-shifting gratings as an inducer, which made it impossible for the observer to discern the inducer's motion direction at the time of the flash and demonstrated that this unpredictable random motion was still able to cause the FDE. These studies suggest that motion information at early visual processing stages can also produce the FDE. 

These accumulated results strongly suggest that the motion information in various stages of visual processing can influence position perception. From a physiological point of view, some researchers have argued that the perceived position shift is caused by feedback signals from extrastriate motion-processing areas, such as the middle temporal area (MT/V5), back to early finely retinotopic areas, such as the primary visual cortex (V1) (Arnold, Thompson, & Johnston, 2007; Durant & Johnston, 2004; Nishida & Johnston, 1999). Consistent with this idea, a few functional magnetic resonance imaging (fMRI) studies have found shifts in activity patterns that may relate to the perceived position shift in early visual areas including V1 (Maus, Fischer, & Whitney, 2009; Whitney et al., 2003). Several physiological studies have shown that from the population-coding perspective, the coded position of the moving object can be shifted in the direction of motion at very early stages in salamander and rabbit retinas (Berry, Brivanlou, Jordan, & Meister, 1999), cat V1 (Fu, Shen, Gao, & Dan, 2004; Jancke, Erlhagen, Schoner, & Dinse, 2004), and monkey V4 (Sundberg, Fallah, & Reynolds, 2006). Thus, the position information distorted by visual motion may be neuronally coded, at least in part, in early finely retinotopic areas. If so, is it possible that this distorted position information further influences the visual processing stage at which adaptable position mechanisms may exist? While several studies have demonstrated that motion adaptation can distort perceived patterns for subsequently viewed stimuli (McGraw, Walsh, & Barrett, 2004; McGraw, Whitaker, Skillen, & Chung, 2002; Nishida & Johnston, 1999; Snowden, 1998; Whitney, 2005; Whitney & Cavanagh, 2003), it is not yet clear whether adaptation to a spatial pattern perceptually distorted by motion can cause any spatial aftereffect. To answer this question, the present study investigated whether the change in the perceived position caused by the FDE contributed to a negative aftereffect in the relative position of two stimuli. 

The position aftereffect (PAE) is an illusion in which, after adaptation to two vertically off-aligned patches, vertically aligned patches appear to be slightly offset in the direction opposite to the physical spatial offset in the adapting stimulus (Hess & Doshi, 1995). A simple linear filtering mechanism cannot explain the PAE because the adaptation occurs even when the carrier orientation of the adapting Gabor patches is orthogonal to the interpatch orientation. Thus, it is likely that the PAE occurs at least after a nonlinear rectification process following the linear filtering mechanism. 

The purpose of the present study was to determine whether the positional alteration corresponding to the FDE influences the visual processing stage at which the mechanism responsible for the PAE is operational. We measured the PAE after adaptation to a patch pair whose offset was perceptually dissociated from the physical one due to the FDE. We used various physical offsets. Thus, under one condition, the physical offset between two patches was set so that physically aligned patches would be perceived as misaligned. Under another condition, they were physically misaligned but were perceived as aligned. If the position information that altered due to the FDE affected the mechanism underlying the PAE, the PAE would be perceived under the former but not the latter condition (thick broken line in Figure 1). In contrast, if the altered position information were independent of the mechanism related to the PAE, the PAE would be perceived under the latter but not the former condition (thick solid line in Figure 1). 

Figure 1

 

Schematics of the predictions. The theoretical PAE strength is plotted against horizontal offset. At the origin, the two patches are physically aligned but perceptually misaligned due to the FDE. To perceive them as aligned, the two patches must be physically misaligned at the offset value, as indicated by the thin broken line. If the position information that was altered due to the FDE affected the mechanism underlying the PAE, the absolute size of the PAE would be minimal when the two patches were physically misaligned but perceived as aligned (thick broken curve). In contrast, if the altered position information was independent of the mechanism related to the PAE, the absolute size of the PAE would be minimal when the two patches were physically aligned (thick solid curve).

Methods

Subjects

One of the authors (TF) and six subjects unaware of the purpose of the experiment (aged 18–27) participated in the study. All had normal or corrected-to-normal visual acuity. Each subject provided written informed consent. This study, which followed the Declaration of Helsinki guidelines, was approved by the Ethics Committee of the College of Arts and Sciences, The University of Tokyo. 

Apparatus

Stimuli were presented in a dark room on a CRT monitor (Mitsubishi Electric RDF223H [Mitsubishi Electric, Tokyo, Japan], 1600 × 1200 pixels, refresh rate 75 Hz, mean luminance 46.3 cd/m²). Each subject placed his or her head on a chin rest and used both eyes to view the stimuli. The viewing distance was 57.7 cm. 

Stimuli

A schematic of the stimulus configuration is shown in Figure 2. A fixation point (a bull's eye) was presented at the center of the display throughout the experiment. Two oppositely moving sinusoidal gratings (both had 0.33 cpd spatial frequency, 3.75°/s velocity, and 50% contrast) were used as the “inducer” of the FDE. The phase of each grating was randomized for every presentation. Each grating occupied a rectangular region subtending 40° × 5° and was centered at 5.5° above and below the central fixation point. A pair of horizontal Gabor patches, hereafter termed the “flash,” was presented between the inducer. Their distance from center to center was 4°, the SD of each envelope was 0.25°, the spatial frequency of each carrier was 1.67 cpd, the contrast was 99%, and the phase of the carrier was randomized for every presentation. The flash pair was presented at a randomly chosen location 5.5°–6.5° to the left of the fixation point. The temporal structure of the adapting stimulus is shown in Figure 3A. In the adaptation period, the motion direction of the inducer was reversed every 1.44 s to avoid motion adaptation. Immediately after the inducer started to move in a certain direction, referred to as the “dragging direction,” the flash pair was flashed three times at the same location at intervals of 0.48 s. To maximize the FDE, the two Gabor patches that belonged to each flash pair were presented asynchronously, with a 160-ms temporal offset. The duration of a single flash was 27 ms (two frames). Adaptation periods starting with or without the “dragging direction” were randomly selected for each presentation. We randomly selected which flash was presented first to each subject. The presentation order determined at the beginning of the experiment was maintained for each subject. 

Figure 2

 

Schematic of the stimulus configuration. Two oppositely moving sinusoidal gratings were used as the inducer of the FDE. The pair of Gabor patches was briefly presented as a flash stimulus. The horizontal position of the midpoint between the two Gabor patches was randomly chosen within the range of 6 ± 0.5° to the left of the fixation point.

Figure 3

 

Time course of experiment. (A) The temporal structure of the adapting and test stimuli. “Upper inducer” and “Lower inducer” indicate the motion directions of the upper and lower inducers, respectively. The periods shown on the darker background indicate the periods of the “dragging direction” within which the flash pair appeared three times. “Upper flash” and “Lower flash” indicate the timing of the upper and lower flashes in reference to the periods of the dragging direction. (B) Schematic of the protocol used to measure the PAE.

Procedure

We examined whether the PAE occurred based on the physical offset or based on the perceived offset between the flash pair. Two dragging directions (i.e., leftward and rightward) and nine physical offsets (i.e., −48, −36, −24, −12, 0, 12, 24, 36, and 48 min; positive numbers indicate that the offset direction was opposite to the direction of the FDE) were possible. The dragging direction and physical offset were chosen from among these options for each session and were consistently used throughout all adaptation periods within each session. Each experimental session was designed to test the PAE after adaptation to the flash pair that was presented many times at a fixed physical offset but that was perceived as having a different offset from the physical one due to the occurrence of the FDE. Each session started with an initial adaptation period including 81 flashes (77.8 s), followed by repeats of test trials intermingled with periods of top-up adaptation (Figure 3B). At the beginning of each trial, the adapting stimulus, which included nine flashes (8.64 s), was presented, serving as top-up adaptation. After a 0.5-s interstimulus interval, a test stimulus was presented for 27 ms (two frames). The test stimulus was the same as the pair of horizontal Gabor patches presented during the adaptation period except that both Gabors were presented at the same time and their contrast was 50%. The test stimulus was followed by a 0.75-s blank period, during which the subject pressed one of two keys to respond to a two-alternative forced-choice question about whether the upper Gabor patch in the test stimulus appeared to the right or to the left relative to the lower one. The next trial started immediately after the 0.75-s blank period regardless of whether the subject pressed a key. 

If the PAE occurred, the pair of Gabor patches in the test stimulus would appear as horizontally offset when they were actually aligned and as aligned when they were actually offset. Using the staircase method, we measured the physical offset between the pair in the test stimulus that was required to establish subjective alignment. In each session, two different staircases were randomly interleaved; in both staircases, the initial offset between the Gabor-patch pair was randomly chosen from the range of ±6 min, and the initial step size was 6 min. The step size was halved after each reversal of the response direction unless it overshot the minimum step size of 1.5 min. Each staircase terminated at the sixth reversal. The point of subjective alignment was determined as the average of the last four reversal points. Each subject completed two sessions under each condition; hence, four estimates (2 staircases × 2 sessions) were obtained under each condition for each subject. We took the average of these estimates as the strength of the PAE. 

In another session, we measured the size of the FDE induced by the adapting stimulus. We determined the physical offset between the flash pair required to establish the perceptual alignment (hereafter called “FDE-canceling point”) for each subject by using the method of constant stimuli. Each session was composed of multiple blocks, and the stimulus was presented in the same way within each block as it was during the top-up adaptation period in the measurement of the PAE, with the exception that the physical offset of the flash pair was not fixed, as explained below. The inducer moved in one direction for 1.44 s and changed its direction five times within each block, which lasted 8.64 s. Immediately after the inducer started to move in a dragging direction, the flash pair was flashed three times at intervals of 0.48 s, with a physical offset that was randomly chosen from the following seven values: −45, −22.5, 0, 22.5, 45, 67.5, and 90 min (positive numbers indicate the offset direction that is opposite to the FDE). After the last flash presentation, subjects judged whether the upper Gabor patch appeared to the right or to the left relative to the lower one by pressing one of two keys presented as a two-alternative forced choice. Because the dragging direction appeared three times within each block, the subject performed the offset-judgment task three times per block. The dragging direction and the initial motion direction were randomly chosen at the beginning of each block. Each subject was tested for three such sessions, and 24 responses were collected under each condition to estimate the probability of “the upper Gabor appearing to the right.” Data collected under the condition in which the upper inducer was moving rightward were flipped and merged with those collected under the condition in which the upper inducer was moving leftward. By fitting a psychometric function in the form of the cumulative Gaussian function, the FDE-canceling point was obtained for each subject. 

Results

A physically upright but perceptually offset adapting flash pair does not cause the PAE

We confirmed that the strength of the PAE varied across the nine flash-pair offsets of the adapting stimulus. The two-way repeated-measures ANOVA with “motion direction of the inducer” and “adapted physical offset between the flash pair” as the two independent variables revealed a significant interaction for each individual, F(8, 54) = 22.59, p < 0.001 (TF); F(8, 54) = 12.98, p < 0.001 (AT); F(8, 54) = 28.71, p < 0.001 (RH); F(8, 54) = 2.29, p < 0.05 (SS); F(8, 54) = 33.03, p < 0.001 (SI); F(8, 54) = 29.76, p < 0.001 (KM); F(8, 54) = 20.47, p < 0.001 (TS), and for the averaged data, F(8, 480) = 88.55, p < 0.001 (average). The significant interaction indicated that the PAE was significant for all subjects given that the sign of the flash-pair offset depended on the motion direction of the inducer. 

Individual and averaged data are shown in Figure 4, in which the magnitude of the PAE is plotted as a function of the physical offset between the flash pair in the adapting stimulus. The error bars indicate ±1 SE. The positive magnitudes in the ordinate indicate that the physical offset between the test Gabor patches that canceled the PAE was rightward (clockwise). In contrast, the positive magnitudes in the abscissa indicate that the physical offset between the adapting Gabor patches was opposite to the direction of the FDE. Thus, under the right-FDE condition, the positive value means leftward, whereas under the left-FDE condition, the positive value means rightward. For the sake of simplicity, in Figure 4, we flipped and merged the data from the right-FDE condition into those from the left-FDE condition and plotted the merged data. The origin of the abscissa indicates the physical vertical alignment between the flash pair during adaptation. Under this condition, the pair was perceptually misaligned due to the FDE dragging the two Gabor patches away from each other. 

Figure 4

 

Individual and averaged PAE data plotted as a function of the physical offset between the flash pair in the adapting stimulus. Error bars indicate ±1 SE. The positive magnitudes in the ordinate indicate that the physical offset between the test Gabor patches that canceled the PAE was rightward (clockwise). The positive magnitudes in the abscissa indicate that the physical offset between the adapting Gabor patches was opposite to the direction of the FDE. Thus, the positive value indicates leftward under the right-FDE condition and rightward under the left-FDE condition. For the sake of simplicity, however, we flipped and merged the data from the right-FDE condition into those from the left-FDE condition and plotted only the merged data. The FDE-canceling point for each individual subject is plotted as a solid vertical line. A shaded area around each FDE-canceling point indicates the 95% bootstrap confidence interval. A solid gray line for each plot indicates the best-fit linear function that accounts for each PAE data point. A vertical broken line for each plot indicates the zero-crossing point of the best-fit model function. A cross-hatched area around the zero-crossing point indicates the 95% bootstrap confidence interval.

Next, we confirmed that the flash pair we used in the adaptation experiment was suitable for yielding the FDE. The FDE-canceling point for each individual subject is plotted in Figure 5 together with the averaged data, with each error bar indicating a 95% bootstrap confidence interval. The FDE was significant with the bootstrap significance level of p < 0.01–0.05 for all subjects but KM, for whom significance was marginal (p = 0.08). 

Figure 5

 

Individual and averaged FDE-canceling points. Error bars on the individual data indicate 95% bootstrap confidence intervals. Asterisks above the bars indicate the bootstrap significance level (*p < 0.05, **p < 0.01).

After confirming the occurrence of a significant PAE and FDE, we proceeded to compare these values on the same chart. In Figure 4, a vertical solid line with a gray-shaded area in each individual plot indicates the FDE-canceling point with a 95% bootstrap confidence interval, taken from Figure 5 and plotted along the same abscissa. This is the very point at which the two Gabor patches were physically shifted to nullify the FDE so that they subjectively appeared to be vertically aligned. Thus, if the PAE occurred based on the perceived offset, the PAE would reverse from negative to positive around this FDE-canceling point. However, both the individual and the averaged data show that the PAE reversed at the origin, that is, at the point of physical alignment, for all but subject TS, for whom only a small FDE was obtained. 

We estimated the point at which the sign of the PAE reversed (zero-crossing point) by fitting a linear function as a first-order approximation because the PAE did not exhibit clear saturation nonlinearity within the range of the adapting stimulus used in this experiment. The model was formulated as: 

 

where s is the slope, r is the zero-crossing point, and x and y correspond to the abscissa and ordinate, respectively, of Figure 4. The fits were reasonably good, with R² = 0.984 for subject TF, 0.877 for AT, 0.992 for RH, 0.728 for SS, 0.910 for SI, 0.945 for KM, 0.943 for TS, and 0.989 for the averaged data. The estimated parameters (s, r) of the best-fit function were (0.28, 0.55 min) for subject TF, (0.18, −2.74 min) for AT, (0.19, −0.50 min) for RH, (0.07, 0.76 min) for SS, (0.17, −2.01 min) for SI, (0.20, −3.85 min) for KM, (0.12, 6.20 min) for TS, and (0.17, 0.00 min) for the averaged data. The estimated slope (s) in the best-fit function indicates that the perceived offset in the PAE was 7%–28% of the physical offset of the adapting stimulus. The estimated zero-crossing point (shown in Figure 4 as a vertical broken line for each plot) was located at the physical alignment for most of the subjects, irrespective of the value of the FDE-canceling point. To further examine the significance of the results, we estimated the 95% confidence intervals of the zero-crossing point for the individual data by bootstrapping; these are shown in Figure 4 as cross-hatched areas around the zero-crossing point. We found that the 95% confidence interval of five of the seven subjects did not overlap with the 95% confidence interval of the FDE-canceling point. It should be noted that the perceived FDEs for the other two subjects (KM and TS) were very small, and the FDE was not even significant for one subject (KM). Therefore, we concluded that the perceived position influenced by the FDE is not effective in inducing the PAE. 

Presentation of the first of the three successive adapting flash pairs alone did not alter the results

During the adaptation period of the previous experiment, we presented the flash pair three times successively after the inducer started to move in the dragging direction to maximize the effect of adaptation. However, presenting the flashes at different times relative to motion onset is thought to increase FDE variance in the adaptation period because FDE size decreases as a function of the onset of the flash relative to that of the inducer's motion (Roach & McGraw, 2009; Whitney & Cavanagh, 2000). It is possible that a large variance in the perceived position eliminated the influence of the FDE on the PAE; thus, we conducted an additional experiment in which the flash pair was presented only once at the onset of each dragging motion during the adaptation period. The inducer reversed its motion direction every 0.61 s; thus, the flash pair was presented at an interval of 1.23 s. The initial adaptation period contained 60 flashes (73.6 s), and each top-up adaptation contained six flashes (7.36 s). We examined the PAE with three rightward physical offsets (0, 24, and 48 min) for the left-FDE condition and three leftward physical offsets (0, 24, and 48 min) for the right-FDE condition. The stimuli and procedure were otherwise the same as those in the main experiment. The size of the FDE in the new configuration was measured in the same fashion as that in the main experiment. One of the authors (TF) and three newly recruited subjects who were unaware of the purpose of the experiment (aged 23–30) participated. The individual results are shown in Figure 6. The one-way repeated-measures ANOVA revealed a significant effect of the six conditions for each individual subject: F(5, 18) = 22.58, p < 0.001 (TF); F(5, 18) = 50.23, p < 0.001 (DH); F(5, 18) = 33.35, p < 0.001 (KI); F(5, 18) = 12.96, p < 0.001 (TO). The test for multiple comparisons revealed significant differences at the physical offsets marked by asterisks in the figure. The overall results were similar to those of the main experiment; a significant PAE occurred as a result of the physical offset. Although subject DH showed a small but significant PAE under the physically aligned condition, it occurred in the direction opposite to that predicted. Thus, the variance in the FDE magnitude that may have been introduced by varying the timing of the onset of the flash relative to that of the onset of the inducer's motion did not significantly affect the results. 

Figure 6

 

Individual PAE data in the single-flash experiment. The PAE magnitude is plotted as a function of the physical offset between the flash pair in the adapting stimulus. Error bars indicate ±1 SE. The data from the right-FDE condition have been flipped and merged with those from the left-FDE condition. Asterisks indicate Bonferroni-corrected multiple comparisons revealing a significant difference between the two left–right symmetrical conditions for that physical offset. The FDE-canceling point for each individual is plotted as a solid vertical line. A shaded area around each FDE-canceling point indicates the 95% bootstrap confidence interval.

It could be argued that even if the flash were presented once at every motion onset, a large inherent variance may exist in the FDE. However, we conducted a preliminary experiment in which the FDE was not induced and the size of the spatial offset between the flash pair was physically jittered during the adaptation period and found that the PAE occurred based on the average size of the adapting physical offset (data not shown). Thus, it is unlikely that a variance in the perceived position eliminated the influence of the FDE on the PAE. 

Asynchronous display of the test flash pair did not alter the results

In the main experiment, the flash pair was presented asynchronously to maximize the FDE during the adaptation period, whereas the test flash pair was presented synchronously. We found no effect of the FDE on the PAE in the synchronous test patches. However, it is possible that any effect of the FDE that occurs in response to asynchronous adapting flashes emerges only when tested with asynchronous flashes. Thus, we conducted an experiment in a subset of spatial offset conditions under which the test flash pair was presented asynchronously using the same temporal offset as that of the flash pair used for the adapting stimulus, while the stimuli and procedure remained the same as those in the main experiment. One of the authors (TF) and three subjects who were unaware of the purpose of the experiment (RH, DH, and SH; aged 22–28) participated. Of the four subjects, two, TF and RH, had participated in the main experiment; TF and DH had participated in the additional single-flash experiment previously described, and SH was newly recruited. We examined the PAE using three rightward physical offsets for the left-FDE condition and three leftward physical offsets for the right-FDE condition. The three offsets were 0, 12, and 24 min for subject SH, and 0, 24, and 48 min for the other three subjects. The individual results are shown in Figure 7. The black squares represent the data obtained using the asynchronous test flashes, whereas the gray circles represent the data from the main experiment. (The main experiment was repeated under the corresponding subset of conditions for the newly participating subject, and the one-way repeated-measures ANOVA revealed a significant effect for each individual subject: F(5, 18) = 42.73, p < 0.001 (DH); F(7, 24) = 14.72, p < 0.001 (SH)). The one-way repeated-measures ANOVA for the new experiment (black squares) revealed a significant effect for three subjects: F(5, 42) = 7.92, p < 0.001 (TF); F(5, 18) = 8.00, p < 0.001 (DH); F(5, 18) = 9.10, p < 0.001 (SH), but not for subject RH: F(5,42) = 1.14, p = 0.354. The Bonferroni-corrected multiple comparisons revealed a significant difference between the two left–right symmetrical conditions. As shown in Figure 7, the results differed across subjects: when the test flashes were presented asynchronously, the PAE for subjects DH and SH was larger than that observed in the main experiment; however, the opposite was true for subject RH, and the results of the two experiments were similar for subject TF. These differences may be explained by the possibility that the mechanism underlying the PAE observed in this study has an effective temporal window that has different widths for different individuals. However, the qualitative characteristics were almost the same between the two experiments. The PAE occurred based on the physical offset between the adapting flash pair for three of the four subjects, and the remaining subject, RH, exhibited null results under all offset conditions. These findings suggest that the FDE does not influence the PAE measured using asynchronous or synchronous test flashes. 

Figure 7

 

Individual PAE data for the asynchronous test condition plotted as a function of the physical offset between the flash pair in the adapting stimulus. Error bars indicate ±1 SE. The data from the right-FDE condition have been flipped and merged with those from the left-FDE condition. The black squares represent the data obtained using asynchronous test flashes, and the gray circles represent the data from the main experiment. The asterisk indicates that the PAE differed significantly between the two left–right symmetrical conditions for that physical offset. The asterisks above the gray circles in the plot of subjects DH and SH indicate a significant difference between the two left–right symmetrical conditions, as revealed by Bonferroni-corrected multiple comparisons. The gray circles in the plots for subjects TF and RH represent replots of the data in Figure 4, and the asterisks indicate a significant difference between the two left–right symmetrical conditions. The FDE-canceling point for each individual subject is plotted as a solid vertical line. The gray-shaded area around the FDE-canceling point indicates the 95% bootstrap confidence interval.

Discussion

Summary of the results

The present study measured the PAE after adaptation to the offset between a pair of Gabor patches, the perceived offset of which was dissociated from the physical offset by the occurrence of the FDE. We used various offset sizes to elicit the PAE and found that the PAE depended solely on the physical offset of the adapting stimulus. Additional experiments demonstrated that the possible variance in the FDE size and the use of synchronous test flashes to measure the PAE were not essential to obtaining the data collected in the main experiment. These results indicate that the position information altered by the FDE does not influence the mechanism underlying the PAE. 

Relationship with previous studies

Previous studies have demonstrated the influence of motion information on position perception. Hayes (2000) showed that an array of perceptually aligned drifting Gabor patches was easier to detect than was an array of physically aligned ones when the perceived positions of moving stimuli were altered due to the MIPS. Li, Khuu, and Hayes (2009) found that background motion influences the perceived shape of illusory contours in Kanizsa's triangle, such that a test Kanizsa figure with concave illusory contours appears to form a regular equilateral triangle when it is placed at the center of globally expanding motion. The contour integration process is often thought to lie in early visual stages, in which local orientation detectors exist (Dakin, 1997; Hess, Hayes, & Field, 2003). May and Hess (2008) suggested that the second-order orientation filter, a possible candidate for the mechanism underlying the PAE, may be responsible for the contour-detection task. Hess and Doshi (1995) suggested a partial correspondence between their results and those reported by Paradiso, Shimojo, and Nakayama (1989), which showed a tilt aftereffect following adaptation to a tilted illusory contour. Thus, the results of Hayes (2000) and Li et al. (2009) differ from our finding of no interaction between the PAE and the motion-induced illusion in position. However, important phenomenal differences exist between the MIPS and the FDE induced by our stimuli. In Hayes' (2000) study, closed contours that contained a moving grating on the inside were presented for a long duration and were perceptually shifted in the motion direction. In Li et al.'s (2009) study, illusory contours superimposed on a moving background were presented for a long duration and were perceptually shifted in the direction of the background motion. These stimulus configurations are radically different from the typical stimulus configuration of the FDE, in which a target stimulus needs to be only briefly flashed and can be separated from a moving stimulus. Additionally, evidence has been compiled in support of the claim that the stimuli for the MIPS can allow early mechanisms to contribute to the illusory position shift. For example, several researchers (Arnold et al., 2007; Chung, Patel, Bedell, & Yilmaz, 2007; Tsui, Khuu, & Hayes, 2007; Whitney et al., 2003) have claimed that a deblurring process or direction-dependent gain-control mechanisms contribute to the MIPS. In a human fMRI study, Whitney et al. (2003) reported activity patterns in early retinotopically organized areas including V1 that may correspond to such a deblurring process. Other researchers have interpreted the results as evidence for a direction-dependent bias in the responsivity of individual neurons (Liu, Ashida, Smith, & Wandell, 2006), although whether such an explanation is consistent with the psychophysical evidence (Arnold et al., 2007; Chung et al., 2007; Whitney et al., 2003) remains unclear. Nonhuman physiological findings suggest that a motion-position interaction qualitatively similar to the MIPS can occur in early visual stages (Berry et al., 1999; Jancke et al., 2004; Sundberg et al., 2006). These mechanisms may contribute to the MIPS and may influence the process related to contour integration. Taken together, previous studies have indicated that the neural mechanisms underlying the MIPS may be located at early stages. In contrast, the large retinal distance between the brief flash and the moving stimulus that is a typical condition for the FDE may be a poor condition for the activation of these early mechanisms. 

In addition to the aforementioned MIPS studies, a few previous FDE studies have obtained results suggesting an interplay between early visual processing and the mechanism underlying the FDE. Kosovicheva et al. (2012) demonstrated that the FDE could distort the spatial tuning of the tilt aftereffect. In their experiment, subjects were adapted to a tilted linear grating within a small disk flashed briefly within a rotating texture. The rotation shifted the apparent position of the adaptor. After adaptation, a brief test grating was presented at one of three locations: the retinal or perceived location of the adaptor or an equidistant control location in the opposite direction. The tilt aftereffect was strongest at the retinal location and tended to be larger at the perceived compared with the control location. This result is consistent with an fMRI study that found shifts in retinotopic activity patterns in early visual areas, including V1, that correspond to the illusory position shift of the FDE (Maus et al., 2009). We believe that several important differences in methodology and effect sizes can readily explain the apparent differences between these findings and our results. First, the tilt aftereffect investigated by Kosovicheva et al. (2012) is qualitatively different from the PAE investigated in the present study. The adapting stimulus used by Kosovicheva et al. consisted of a fine tilted luminance-defined grating that activated the first-order orientation mechanism deemed important for the generation of the conventional tilt aftereffect (e.g., Blakemore & Nachmias, 1971; Campbell & Maffei, 1971; Maffei, Fiorentini, & Bisti, 1973; Movshon & Lennie, 1979; Ware & Mitchell, 1974). In contrast, the first-order orientation-selective mechanism was not activated by our adapting or test stimulus (Figure 2) because the luminance carrier of the Gabor patches was always horizontal, and the upper and lower patches were separated by 4° or more, which is too wide to activate a single orientation mechanism. Second, the FDE in the study conducted by Kosovicheva et al. (2012) produced only a slight distortion of the spatial tuning of the tilt aftereffect toward the inducer's motion direction (the size of the normalized tilt aftereffect at the perceived location minus that at the control location was approximately 0.15 when calculated as a fraction of the aftereffect amplitude measured at the retinal location), which may not have been effective enough to alter the behavior of the position mechanism for the PAE. Third, the adapting flashed stimuli used by Kosovicheva et al. (2012) were presented within the rotating texture. In contrast, our adapting flash pair was presented at positions that were separate from those of the moving inducers. As previously mentioned, the retinal distance between the flash and the moving stimulus may be an important factor in activating the early mechanisms that induce motion-position interactions. Finally, the aim of the research differed in the two studies. Kosovicheva et al. (2012) investigated whether the tilt aftereffect within one small physical location transfers to a location where the adapting stimulus does not physically exist but appears to be present as a result of the FDE. They found that the tilt aftereffect was strongest at the physically adapted location, but observed a small bias in the illusion strength in favor of the perceived-location hypothesis. Our study investigated whether two vertical positions separated by 4° and perceptually off-aligned by the FDE can comprise an effective spatial offset signal for adapting the position mechanism and for later making two physically vertical positions appear off-aligned in the opposite direction. Our results indicated that this was not the case. 

Mechanisms responsible for the PAE

Previous investigators have assumed that higher-order elongated orientation filters explain data related to the perception of the relative position of well-separated and nonabutting items (e.g., Hess & Doshi, 1995; Keeble & Nishida, 2001; Kobayashi, Terao, & Murakami, 2012; Whitaker & Latham, 1997). In this model, each item (patch) is detected by first-order luminance filters tuned to properties specific to the item, such as orientation and spatial frequency. The signals from the first-order filters then enter a rectifier followed by a second-stage filter of a larger size that encompasses two regions occupied by the two items. Thus, theoretically, the adaptation of this second-stage filter is a possible candidate for the mechanism mediating the PAE observed in the present study. However, involvement of a simple second-order orientation filter based on filter-rectify-filter operation is unlikely because the PAE persisted when the distance between the two patches was too wide to be processed by a single orientation filter (see Burbeck, 1987; Morgan & Ward, 1985), suggesting that another mechanism tuned for a spatial offset between separate locations must underlie the PAE. Furthermore, it is possible that adaptation occurs not only to the relative position between the flash pair, but also to the absolute position of each component of the flash pair. Whitaker, McGraw, and Levi (1997) demonstrated that adaptation to contrast-defined stimuli containing horizontally biased luminance components could cause an aftereffect on the perceived location of subsequently viewed stimuli in the direction away from the bias direction of the adapting stimulus. In this case, the adaptation might occur at or before the processing stage where the position signals from two independent items are compared. However, the phenomenon reported by Whitaker et al. (1997) alone cannot explain the occurrence of the PAE in the present study because it cannot explain why our horizontally symmetric Gabor-patch stimuli produce a subsequent aftereffect of horizontal spatial offset in the observed direction. 

Conclusion

The present study measured the PAE after adaptation to the offset between a pair of Gabor patches, whose perceived offset was dissociated from the physical offset by the FDE. We used various offset sizes to elicit the PAE and found that the PAE was based on the physical offset of the adapting stimulus. Despite previous physiological studies that suggested interactions between motion and position information in very early retinotopic areas, our results indicate that the internal representations of positions altered by the presence of nearby motion signals do not contribute to the mechanism underlying the PAE. 

Acknowledgments

This study was supported by the Nissan Science Foundation and JSPS Funding Program for Next Generation World-Leading Researchers (LZ004). 

Commercial relationships: none. 

Corresponding author: Taiki Fukiage. 

Email: [e-mail]. 

Address: Department of Life Sciences, University of Tokyo, Meguro-ku, Tokyo, Japan.