Journal of Experimental Psychology: Human Perception and Performance 2014, Vol. 40, No. 3, 938-947
© 2014 American Psychological Association 0096-i523/14/$12.00 DOI: IO.lO37/aOO35362
Illusory Motion Reversals and Feature Tracking Analyses of Movement Derek H, Arnold, Samuel L. Pearce, and Welber Marinovic The University of Queensland Illusory motion reversals (IMRs) can happen when looking at a repetitive pattern of motion, such as a spinning wheel. To date these have been attributed to either a form of motion aftereffect seen while viewing a moving stimulus or to the visual system taking discrete perceptual snapshots of continuous input. Here we present evidence that we argue is inconsistent with both proposals. First, we show that IMRs are driven by the adaptation of nondirectional temporal frequency tuned cells, which is inconsistent with the motion aftereffect account. Then we establish that the optimal frequency for inducing IMRs differs for color and luminance defined movement. These data are problematic for any account based on a constant rate of discrete perceptual sampling. Instead, we suggest IMRs result from a perceptual rivalry involving discrepant signals from a feature tracking analysis of movement and motion-energy based analyses. We do not assume that feature tracking relies on a discrete sampling of input at a fixed rate, but rather that feature tracking can (mis)match features at any rate less than a stimulus driven maximal resolution. Consistent with this proposal, we show that the critical frequency for inducing IMRs is dictated by the duty cycle of salient features within a moving pattern, rather than by the temporal frequency of luminance changes. Keywords: motion perception, illusory motion reversal, attention-based motion perception, adaptation, temporal frequency
Repetitive patterns of motion can sometimes seem to reverse. For instance, if discrete perceptual snap shots of a spinning wheel are taken, as can happen when viewing stimuli lit by a fiickering light source, the most proximate spoke images in successive snap shots might not represent the same spoke. Instead, rotation can be such that each spoke nearly reaches the position occupied by the advanced spoke in the previous snapshot. Consequently, the wheel can seem to spin in reverse. This is called temporal aliasing (see Bach et al., 1997). Not all illusory motion reversals (IMRs) can readily be attributed to physical temporal aliasing, as they can also happen while viewing a rotating pattern lit by a continuous light source, such as the sun (Kline, Holcotnbe, & Eagleman, 2004; Purves, Paydarfar, & Andrews, 1996; VanRullen, Reddy, & Koch, 2005). This has prompted the proposal that discrete sampling inight arise within the human visual system, with discrete perceptual snapshots taken at a rate of ~ 13 Hz (Andrews & Purves, 2005; Purves et al., 1996; VanRullen, 2007; VanRullen et al., 2005; VanRullen, Reddy. & Koch, 2006). Hypothetically, movement suggested by discrete
perceptual snapshots engage in a form of perceptual rivalry with analyses of movement conducted by other, presumably nondiscrete, mechanisms (VanRullen, 2007; VanRullen et al., 2005). An alternate possibility is that IMRs constitute a form of perceptual rivalry between an adaptation-induced motion aftereffect signal in the opposite direction and a persistent representation of movement in the veridical direction (Holcomhe, Clifford, Eagleman, & Pakarian, 2005; Kline, Holcombe, & Eagleman, 2004, 2006). Note that the point of difference between these two explanations of IMRs is the source of the erroneous signal. According to one view it is derived from taking discrete perceptual snapshots at a more or less fixed rate, while according to the other account the erroneous signal is a motion-aftereffect seen while still viewing the adaptor. Here we will present evidence that we argue is inconsistent with the two existing explanations of IMR. First, we will show that IMRs are linked to adaptation, but the requisite adaptation is not direction-tuned. This discredits the suggestion that IMRs result from a motion-aftereffect signal. Second, we will establish that the optimal rate of rotation for IMRs differs systematically for motion signaled by changes in luminance and for movement signaled by color changes. This is problematic for any proposal linking IMRs to a fixed, intrinsic or extrinsically generated, rate of discrete perceptual sampling (see Discussion for more coverage of this theory).
This article was published Online First March 17, 2014. Derek H. Arnold, Samuel L. Pearce, and Welber Marinovic, School of Psychology, The University of Queensland, St. Lucia, Queensland, Australia. This work was supported by an Australian Research Council Discovery Early Career Research Award to Weber Maririovic and by a Discovery Project Grant and Australian Research Fellowship to Derek H. Arnold. We thank Dr. Rufin VanRullen, Dr. John Cass, and Assoc. Prof. Alex Holcombe for feedback concerning this project. Correspondence concerning this article should be addressed to Derek H. Arnold, School of Psychology, The University of Queensland, Australia. E-mail: [email protected]
Experiment 1: IMRs Are Driven by Adaptation of Nondirectional Temporal Frequency Tuned
Mechanisms A link between IMRs and neural adaptation seems clear as IMRs tend not to be reported until a persistent moving stimulus has been viewed for some time (> —13 s, see Kline et al., 2004). However, it is unclear what the optimal adaptation stimulus is for 938
ILLUSORY MOTION REVERSALS
inducing IMRs. To date there has been an implicit assumption that the optimal adapting stimulus is also the optimal test stimulus, namely a pattern rotating at a rate that generates —10-15 Hz repetitions of the moving pattern (Andrews & Purves, 2005; Purves et al., 1996; VanRullen, 2007; VanRullen et al., 2005, 2006). However, no previous study has attempted to use different adaptor and test frequencies to induce IMRs, so it is unclear if the optimal test frequency is also the optimal adaptor frequency. Here we test this assumption, and find it wanting.
Method Five volunteers participated in Experiment 1, including the flrst two authors. All had normal or corrected-to-normal visual acuity. All participants completed the Same Direction adaptation experimental condition, whereas four (one author and three volunteers) completed the Opposite Direction adaptation condition (see below). Visual stimuli were generated with a ViSaGe stimulus generator from Cambridge Research Systems (Rochester, United Kingdom), programmed in MATLAB 7.5 and displayed on a gammacorrected 19 in. Sony Trinitron G420 monitor at a resolution of 1,024 X 768 pixels and a refresh rate of 120 Hz. Stimuli were viewed from 57 cm with the head placed in a chin rest. The display background was gray (CIE, Gray; x = 0.28, y = 0.33, Y = 35.52) and participant responses were reported via mouse button presses. Adapting and test stimuli consisted of an array of four sinusoidal gratings with a spatial frequency of 2 cycles/degree of visual angle (dva), all drifting at identical rates (see Figure 1). Each of the four square gratings subtended 3.9 dva and were centered 4.9 dva to the
15 sees
0.5 sees
= ¿
\ 5 sees
Time Figure 1. Time course of events in Experiment la. Arrows indicate drift directions for CW rotation and were not shown during the experiments.
939
left, right, above and below a central red fixation point. Gratings above and below fixation were vertical whereas those to the left and right of fixation were horizontal. The waveforms of adapting and test stimuli drifted in directions somewhat consistent with either a clockwise (CW) or counter clockwise (CCW) rotation. Each trial consisted of 15 s of adaptation, followed by a 500 ms interstimulus-interval, and then a 5 s test stimulus presentation (see Figure 1). At the end of each trial the participant reported if all test elements had seemed to simultaneously reverse during the test presentation by pressing one of two response buttons. Participants were asked not to report instances of ambiguous motion or lack of motion. Each block of trials consisted of 50 individual trials. Participants completed fotir blocks of trials for each of six adapting temporal frequencies (1, 2.5, 5, 10, 20, and 30 Hz). Tests were always animated at 10 Hz. During a block of trials adapting stimuli rotated in the same direction and tests either rotated in the same (25 trials) or in the opposite direction (25 trials). Completing four blocks of trials for each TF allowed us to sample each combination of adaptor and test direction. Blocks of trials were separated by breaks of at least half an hour, and were usually completed on separate days. Each participant completed 24 blocks of trials, encompassing 1,200 individual trials.
Results The proportion of motion reversals indicated a quadratic trend (F, 4 = 47.46, p = .002, y\l = 0.92). We can describe these results in terms of there being a steady decrease in the proportion of IMRs until a minimum was reached for 20 Hz adaptors, then there was an increase for 30 Hz adaptors. The pattern of results for tests and adaptors moving in opposite directions was similar (see Figure 2B), resulting in a near significant quadratic trend (F, 3 = 9.25, p = .056, -4 = 0.75). If we compare the results of adapting to the same and opposite directions of motion, we find that there is no main effect for direction of adaptation (F; , = 0.95, p = .40, % = 0.24). There is, however, a significant main effect for temporal frequency (F, ,5 = 3.19, p = .02, Tip = 0.56). More important, there is no evidence for an interaction between direction of adaptation and temporal frequency (F5 ,5 = 1.42, p = .27, Tij = 0.32). Thus, the results from same and opposite direction adaptation speak to a common cause for the main effect of temporal frequency.
Experiment 2: Illusory Direction Reversals After Adaptation to Directionless Flicker The results of Experiment 1 are inconsistent with directional tuning, as the same qualitative pattern of results was obtained after adapting to the same and opposite directions of motion relative to tests. These data therefore suggest that adapting to nondirectional flicker might have a qualitatively similar impact. Details for Experiment 2 were as for Experiment 1, with the following exceptions. Both adapting and test stimuli consisted of a wagon wheel pattern presented within an annulus and centered on fixation (see Figure 3). The diameter of the outer annulus subtended 13 dva and the inner 10 dva. The wagon wheel pattern was defined by a sinusoidal luminance modulation with a radial frequency of 10 and a Michelson contrast of 100%. During adaptation
940
ARNOLD, PEARCE, AND MARINOVIC 1.0
I.«
Same direction of adaptor
Opposite direetion of adaptor
S O.g
I 0.6
"I 0.6
I
0.4
i
0.4 0.2
. 0.2
0.0
0.0 2.5
5.0
lO.O
2.5 5.0 lo.o ao.o Temporal Frequeney (Hz)
20.0
Temporal Frequency (Hz)
Figure 2. Proportion of trials in which the participants reported illusory motion reversals as a function of adaptor temporal frequency. Data is shown for adaptors moving in the same direction as tests (A) and for adaptors moving in the opposite direction (B). Error bars show ± 1 SEM.
the polarity of the wagon wheel was sinusoidally counter phased at rates of either 1, 2.5, 5, 10, 20, or 30 Hz to generate luminance flicker. During tests the radial grating was modulated to generate counterclockwise rotation at a rate of 1 revolution per second, which generated localized 10 Hz repetitions of the stimulus pattern. Participants completed six blocks of trials, with each block containing 20 trials with repeated exposure to just one of the six adaptation temporal frequencies.
15 sees
Results and Discussion Results for Experiment 2 were qualitatively similar to Experiment 1 (see Figure 4). IMRs were observed for a greater proportion of tests following adaptation to low (
© 2014 American Psychological Association 0096-i523/14/$12.00 DOI: IO.lO37/aOO35362
Illusory Motion Reversals and Feature Tracking Analyses of Movement Derek H, Arnold, Samuel L. Pearce, and Welber Marinovic The University of Queensland Illusory motion reversals (IMRs) can happen when looking at a repetitive pattern of motion, such as a spinning wheel. To date these have been attributed to either a form of motion aftereffect seen while viewing a moving stimulus or to the visual system taking discrete perceptual snapshots of continuous input. Here we present evidence that we argue is inconsistent with both proposals. First, we show that IMRs are driven by the adaptation of nondirectional temporal frequency tuned cells, which is inconsistent with the motion aftereffect account. Then we establish that the optimal frequency for inducing IMRs differs for color and luminance defined movement. These data are problematic for any account based on a constant rate of discrete perceptual sampling. Instead, we suggest IMRs result from a perceptual rivalry involving discrepant signals from a feature tracking analysis of movement and motion-energy based analyses. We do not assume that feature tracking relies on a discrete sampling of input at a fixed rate, but rather that feature tracking can (mis)match features at any rate less than a stimulus driven maximal resolution. Consistent with this proposal, we show that the critical frequency for inducing IMRs is dictated by the duty cycle of salient features within a moving pattern, rather than by the temporal frequency of luminance changes. Keywords: motion perception, illusory motion reversal, attention-based motion perception, adaptation, temporal frequency
Repetitive patterns of motion can sometimes seem to reverse. For instance, if discrete perceptual snap shots of a spinning wheel are taken, as can happen when viewing stimuli lit by a fiickering light source, the most proximate spoke images in successive snap shots might not represent the same spoke. Instead, rotation can be such that each spoke nearly reaches the position occupied by the advanced spoke in the previous snapshot. Consequently, the wheel can seem to spin in reverse. This is called temporal aliasing (see Bach et al., 1997). Not all illusory motion reversals (IMRs) can readily be attributed to physical temporal aliasing, as they can also happen while viewing a rotating pattern lit by a continuous light source, such as the sun (Kline, Holcotnbe, & Eagleman, 2004; Purves, Paydarfar, & Andrews, 1996; VanRullen, Reddy, & Koch, 2005). This has prompted the proposal that discrete sampling inight arise within the human visual system, with discrete perceptual snapshots taken at a rate of ~ 13 Hz (Andrews & Purves, 2005; Purves et al., 1996; VanRullen, 2007; VanRullen et al., 2005; VanRullen, Reddy. & Koch, 2006). Hypothetically, movement suggested by discrete
perceptual snapshots engage in a form of perceptual rivalry with analyses of movement conducted by other, presumably nondiscrete, mechanisms (VanRullen, 2007; VanRullen et al., 2005). An alternate possibility is that IMRs constitute a form of perceptual rivalry between an adaptation-induced motion aftereffect signal in the opposite direction and a persistent representation of movement in the veridical direction (Holcomhe, Clifford, Eagleman, & Pakarian, 2005; Kline, Holcombe, & Eagleman, 2004, 2006). Note that the point of difference between these two explanations of IMRs is the source of the erroneous signal. According to one view it is derived from taking discrete perceptual snapshots at a more or less fixed rate, while according to the other account the erroneous signal is a motion-aftereffect seen while still viewing the adaptor. Here we will present evidence that we argue is inconsistent with the two existing explanations of IMR. First, we will show that IMRs are linked to adaptation, but the requisite adaptation is not direction-tuned. This discredits the suggestion that IMRs result from a motion-aftereffect signal. Second, we will establish that the optimal rate of rotation for IMRs differs systematically for motion signaled by changes in luminance and for movement signaled by color changes. This is problematic for any proposal linking IMRs to a fixed, intrinsic or extrinsically generated, rate of discrete perceptual sampling (see Discussion for more coverage of this theory).
This article was published Online First March 17, 2014. Derek H. Arnold, Samuel L. Pearce, and Welber Marinovic, School of Psychology, The University of Queensland, St. Lucia, Queensland, Australia. This work was supported by an Australian Research Council Discovery Early Career Research Award to Weber Maririovic and by a Discovery Project Grant and Australian Research Fellowship to Derek H. Arnold. We thank Dr. Rufin VanRullen, Dr. John Cass, and Assoc. Prof. Alex Holcombe for feedback concerning this project. Correspondence concerning this article should be addressed to Derek H. Arnold, School of Psychology, The University of Queensland, Australia. E-mail: [email protected]
Experiment 1: IMRs Are Driven by Adaptation of Nondirectional Temporal Frequency Tuned
Mechanisms A link between IMRs and neural adaptation seems clear as IMRs tend not to be reported until a persistent moving stimulus has been viewed for some time (> —13 s, see Kline et al., 2004). However, it is unclear what the optimal adaptation stimulus is for 938
ILLUSORY MOTION REVERSALS
inducing IMRs. To date there has been an implicit assumption that the optimal adapting stimulus is also the optimal test stimulus, namely a pattern rotating at a rate that generates —10-15 Hz repetitions of the moving pattern (Andrews & Purves, 2005; Purves et al., 1996; VanRullen, 2007; VanRullen et al., 2005, 2006). However, no previous study has attempted to use different adaptor and test frequencies to induce IMRs, so it is unclear if the optimal test frequency is also the optimal adaptor frequency. Here we test this assumption, and find it wanting.
Method Five volunteers participated in Experiment 1, including the flrst two authors. All had normal or corrected-to-normal visual acuity. All participants completed the Same Direction adaptation experimental condition, whereas four (one author and three volunteers) completed the Opposite Direction adaptation condition (see below). Visual stimuli were generated with a ViSaGe stimulus generator from Cambridge Research Systems (Rochester, United Kingdom), programmed in MATLAB 7.5 and displayed on a gammacorrected 19 in. Sony Trinitron G420 monitor at a resolution of 1,024 X 768 pixels and a refresh rate of 120 Hz. Stimuli were viewed from 57 cm with the head placed in a chin rest. The display background was gray (CIE, Gray; x = 0.28, y = 0.33, Y = 35.52) and participant responses were reported via mouse button presses. Adapting and test stimuli consisted of an array of four sinusoidal gratings with a spatial frequency of 2 cycles/degree of visual angle (dva), all drifting at identical rates (see Figure 1). Each of the four square gratings subtended 3.9 dva and were centered 4.9 dva to the
15 sees
0.5 sees
= ¿
\ 5 sees
Time Figure 1. Time course of events in Experiment la. Arrows indicate drift directions for CW rotation and were not shown during the experiments.
939
left, right, above and below a central red fixation point. Gratings above and below fixation were vertical whereas those to the left and right of fixation were horizontal. The waveforms of adapting and test stimuli drifted in directions somewhat consistent with either a clockwise (CW) or counter clockwise (CCW) rotation. Each trial consisted of 15 s of adaptation, followed by a 500 ms interstimulus-interval, and then a 5 s test stimulus presentation (see Figure 1). At the end of each trial the participant reported if all test elements had seemed to simultaneously reverse during the test presentation by pressing one of two response buttons. Participants were asked not to report instances of ambiguous motion or lack of motion. Each block of trials consisted of 50 individual trials. Participants completed fotir blocks of trials for each of six adapting temporal frequencies (1, 2.5, 5, 10, 20, and 30 Hz). Tests were always animated at 10 Hz. During a block of trials adapting stimuli rotated in the same direction and tests either rotated in the same (25 trials) or in the opposite direction (25 trials). Completing four blocks of trials for each TF allowed us to sample each combination of adaptor and test direction. Blocks of trials were separated by breaks of at least half an hour, and were usually completed on separate days. Each participant completed 24 blocks of trials, encompassing 1,200 individual trials.
Results The proportion of motion reversals indicated a quadratic trend (F, 4 = 47.46, p = .002, y\l = 0.92). We can describe these results in terms of there being a steady decrease in the proportion of IMRs until a minimum was reached for 20 Hz adaptors, then there was an increase for 30 Hz adaptors. The pattern of results for tests and adaptors moving in opposite directions was similar (see Figure 2B), resulting in a near significant quadratic trend (F, 3 = 9.25, p = .056, -4 = 0.75). If we compare the results of adapting to the same and opposite directions of motion, we find that there is no main effect for direction of adaptation (F; , = 0.95, p = .40, % = 0.24). There is, however, a significant main effect for temporal frequency (F, ,5 = 3.19, p = .02, Tip = 0.56). More important, there is no evidence for an interaction between direction of adaptation and temporal frequency (F5 ,5 = 1.42, p = .27, Tij = 0.32). Thus, the results from same and opposite direction adaptation speak to a common cause for the main effect of temporal frequency.
Experiment 2: Illusory Direction Reversals After Adaptation to Directionless Flicker The results of Experiment 1 are inconsistent with directional tuning, as the same qualitative pattern of results was obtained after adapting to the same and opposite directions of motion relative to tests. These data therefore suggest that adapting to nondirectional flicker might have a qualitatively similar impact. Details for Experiment 2 were as for Experiment 1, with the following exceptions. Both adapting and test stimuli consisted of a wagon wheel pattern presented within an annulus and centered on fixation (see Figure 3). The diameter of the outer annulus subtended 13 dva and the inner 10 dva. The wagon wheel pattern was defined by a sinusoidal luminance modulation with a radial frequency of 10 and a Michelson contrast of 100%. During adaptation
940
ARNOLD, PEARCE, AND MARINOVIC 1.0
I.«
Same direction of adaptor
Opposite direetion of adaptor
S O.g
I 0.6
"I 0.6
I
0.4
i
0.4 0.2
. 0.2
0.0
0.0 2.5
5.0
lO.O
2.5 5.0 lo.o ao.o Temporal Frequeney (Hz)
20.0
Temporal Frequency (Hz)
Figure 2. Proportion of trials in which the participants reported illusory motion reversals as a function of adaptor temporal frequency. Data is shown for adaptors moving in the same direction as tests (A) and for adaptors moving in the opposite direction (B). Error bars show ± 1 SEM.
the polarity of the wagon wheel was sinusoidally counter phased at rates of either 1, 2.5, 5, 10, 20, or 30 Hz to generate luminance flicker. During tests the radial grating was modulated to generate counterclockwise rotation at a rate of 1 revolution per second, which generated localized 10 Hz repetitions of the stimulus pattern. Participants completed six blocks of trials, with each block containing 20 trials with repeated exposure to just one of the six adaptation temporal frequencies.
15 sees
Results and Discussion Results for Experiment 2 were qualitatively similar to Experiment 1 (see Figure 4). IMRs were observed for a greater proportion of tests following adaptation to low (
