Interactions between concentric form-from-structure and face perception revealed by visual masking but not adaptation

Abstract

Abstract

Abstract:

Abstract  Findings from diverse subfields of vision research suggest a potential link between high-level aspects of face perception and concentric form-from-structure perception. To explore this relationship, typical adults performed two adaptation experiments and two masking experiments to test whether concentric, but not nonconcentric, Glass patterns (a type of form-from-structure stimulus) utilize a processing mechanism shared by face perception. For the adaptation experiments, subjects were presented with an adaptor for 5 or 20 s, prior to discriminating a target. In the masking experiments, subjects saw a mask, then a target, and then a second mask. Measures of discriminability and bias were derived and repeated measures analysis of variance tested for pattern-specific masking and adaptation effects. Results from Experiment 1 show no Glass pattern-specific effect of adaptation to faces; results from Experiment 2 show concentric Glass pattern masking, but not adaptation, may impair upright/inverted face discrimination; results from Experiment 3 show concentric and radial Glass pattern masking impaired subsequent upright/inverted face discrimination more than translational Glass pattern masking; and results from Experiment 4 show concentric and radial Glass pattern masking impaired subsequent face gender discrimination more than translational Glass pattern masking. Taken together, these findings demonstrate interactions between concentric form-from-structure and face processing, suggesting a possible common processing pathway.

Introduction

Face perception involves processing the arrangements of features

Humans are able to perceive and discriminate other human faces rapidly (Lehky, 2000; Tsao & Livingstone, 2008), despite their visual complexity. The processing mechanisms involved in face discrimination are unclear. Some argue that face discrimination is domain-specific (Kanwisher, 2000; Tsao & Livingstone, 2008), in that other types of stimuli do not utilize key processing mechanisms involved in face perception. Others argue that face discrimination is domain-general: A set of processing mechanisms is utilized in the development of visual expertise and becomes involved in face perception during development (Diamond & Carey, 1986; Gauthier & Nelson, 2001). Despite these disagreements, many studies suggest that the rapid perception of a face, which is necessary for face discrimination, involves processing the arrangements of the features of a face. For example, face parts are better recognized when a whole upright face is presented (Rhodes, Brake, & Atkinson, 1993; Tanaka & Farah, 1993). Distorted face features are more easily noticed when a face is upright than when it is upside down (Boutsen, Humphreys, Praamstra, & Warbrick, 2006; Lewis & Johnston, 1997; Thompson, 1980). Face discrimination is impaired when two halves of different faces are aligned with one another (Maurer, Grand, & Mondloch, 2002; Mondloch & Maurer, 2008; Young, Hellawell, & Hay, 1987). One might conclude from this work that inverted and upright faces are processed using different mechanisms. However, it is possible that both inverted and upright faces utilize the same set of processing mechanisms, and that the difference in upright and inverted face perception relates to processing efficiency (Sekuler, Gaspar, Gold, & Bennett, 2004). 

Understanding the processing of face feature arrangements may be enhanced by studying simpler stimuli

Glass patterns are a simpler class of stimuli that also involve the processing of arrangements of features. A Glass pattern is constructed by a set of dot pairs, called dipoles (Glass, 1969). Each dipole has one dot randomly placed within the pattern, and a member dot that is displaced from the random dot according to a geometric rule (see Wilson & Wilkinson, 1998, for equations). The geometric rule defines the global form emerging from the dipoles, which are meaningless at the elemental level. A rotational rule produces a concentric pattern. Orienting the dipoles along imaginary radii from a common center produces a radial pattern. Shifting the member dot vertically or horizontally produces a translational pattern. 

Concentric Glass pattern perception may utilize a mechanism that is important for face perception

Psychophysical, physiological, and human lesion data have hinted that concentric Glass pattern perception may utilize a mechanism shared by face perception. As with faces, concentric and radial, but not translational, Glass pattern perception improves when the whole pattern is presented, and degrades as the form is restricted to a piece of the pattern (Wilson & Wilkinson, 1998; Wilson, Wilkinson, & Asaad, 1997). This psychophysical data formed the basis for a biologically plausible model of concentric Glass pattern perception that uses contour information from head shape to identify symmetry in natural face photographs (Poirier & Wilson, 2010); head shape has been shown to be important in processing head orientation and viewpoint (Wilson & Wilkinson, 2002). We propose that a generalized version of their model may use the arrangement of the face features, and that these processes may operate synergistically. Both types of processing support one another and, together, may enhance the ability to process symmetry in faces (Or & Wilson, 2013; Poirier & Wilson, 2010). 

Other psychophysical data have shown that concentric, but not translational, Glass patterns are more easily and quickly processed when the pattern size increases from 3° to 10°. We speculate that neurons with larger receptive field sizes, such as those found in V4, may be utilized in the perception of concentric, but not translational, Glass patterns (Aspell, Wattam-Bell, & Braddick, 2006). Monkey electrophysiological data show that neurons in V1 and V2 are not tuned for concentric Glass patterns (Smith, Kohn, & Movshon, 2007). Human imaging studies show that BOLD responses from successively higher level visual regions better classify concentric, but not translational, Glass patterns (Ostwald, Lam, Li, & Kourtzi, 2008). Midlevel visual regions, such as V3 and possibly V4, show increased BOLD signal for coherent concentric and radial Glass patterns (Mannion, Kersten, & Olman, 2013). The BOLD signal increases for concentric forms but not other forms in face-sensitive brain regions (Wilkinson et al., 2000). Acquired prosopagnosia, but not acquired alexia, may also impair concentric Glass pattern perception (Rentschler, Treutwein, & Landis, 1994). 

Visual masking and adaptation may be used to test this hypothesis directly

While the above studies have hinted at a common processor for concentric Glass patterns and faces, to our knowledge, no study has directly tested whether concentric Glass patterns and the arrangement of face features engage a common processing mechanism. Although a functional MRI (fMRI) study has shown that face-specific regions respond to concentric Glass patterns (Wilkinson et al., 2000), such evidence does not prove that a common processing mechanism may be engaged. Evidence of behavioral interactions between concentric Glass patterns and faces would provide strong convergent evidence that Glass patterns and faces may share a common processing mechanism. Therefore, we sought to test this hypothesis using visual adaptation and masking paradigms. Visual adaptation occurs when one views a stimulus, an adaptor, for a long period of time (Clifford et al., 2007; Kohn, 2007; Krekelberg, Boynton, & van Wezel, 2006). The viewing alters subsequent processing to create an aftereffect: A subsequently presented stimulus, the target, is perceived differently. Physiologically, visual adaptation may result from the effect of the adaptor on individual or populations of neurons that are responsive to the target. The physiological effects of visual adaptation can be tested via fMRI paradigms (Krekelberg et al., 2006). Visual masking occurs when a stimulus, the mask, is presented either before or after the appearance of a target. The presentation of the mask limits or enhances the perception of the target by interfering with the processing of the target. It has been shown that arrangements of face features and face outlines, but not houses, can mask the presentation of whole faces (Loffler, Gordon, Wilkinson, Goren, & Wilson, 2005). Therefore, concentric Glass patterns may share a common processing mechanism with faces if concentric Glass pattern masking of faces impairs face perception. Operationally, visual masking and visual adaptation can be dissociated behaviorally by examining the duration of the effects. Effects of visual masking typically last less than 150 ms (Breitmeyer, 2007; Breitmeyer & Ogmen, 2000; Rolls, 2004), while effects of visual adaptation can last more than 500 ms (Krekelberg et al., 2006). If adaptation to concentric Glass patterns specifically impairs face perception and vice versa, or if concentric Glass pattern masking specifically impairs subsequent face perception, this would suggest that the arrangement of face features and concentric Glass pattern perception utilize a common processor. 

General methods

Participants

Eighteen- to thirty-year-old, right-handed participants were recruited from the local community (see Acknowledgments) and financially compensated for their time. They consented according to procedures approved by the Washington University Human Research Protection Office. Participants had no first-degree relatives with an autism spectrum disorder or attention-deficit/hyperactivity disorder. Participants were screened out for any neurological deficit, strabismus, or vision not correcting to normal acuity with glasses. 

Experimental procedures

Experimental setup

Participants were seated in a darkened room, 53.4 cm away from the center of the monitor. Participants were provided with a chin rest and head strap to minimize head movements (Headspot, Tall Option: University of Houston College of Optometry), and the monitor height was modified using an adjustable stand to ensure that each participant's eyes were level with the center of the monitor. Participants responded to stimuli that appeared on the screen using a Carnegie Mellon University (CMU) button box (New Micros, Inc.; Dallas, TX; 1 ms timing resolution). 

All stimuli were presented on a 17-in. CRT monitor (Dell model E771A, 15.5-in. viewable). The monitor was controlled by a MacBook Pro laptop (Apple Computer, Inc.), and the monitor resolution was set to 1024 × 768 at 75 Hz. All experiments were written in MATLAB (The Mathworks, Inc. Natick, MA), using the Psychophysics Toolbox extensions (Brainard, 1997). The timing of the stimulus presentation was tested using a photodiode connected to a separate Macintosh computer running PowerLab (ADInstruments, Colorado Springs, CO, http://www.adinstruments.com). The luminance of the monitor and all stimuli were measured using a LS-100 photometer (Konica Minolta, Osaka, Japan, http://sensing.konicaminolta.us/). Every possible luminance value for the display setup was measured to create a gamma lookup table to linearize luminance values. The luminance of the monitor could vary from 0 to 69 cd/m². For all experiments, the display background was set to a mean of 22 cd/m². 

Stimulus design

Glass patterns:

Glass patterns were also created with code written in MATLAB, using the Psychophysics Toolbox extensions. Glass patterns were constructed by randomly placing square dots, each with a length of 1.8 minutes of arc, within the extent of an imaginary ellipse. For each randomly placed dot, a second paired dot was displaced 7.3 minutes of arc from the first dot. Dot density was 33% for every pattern. The geometric rule governing the displacement of dots determined the global form of the Glass pattern (Wilson & Wilkinson, 1998). An expansion rule defined the radial forms (Figure 1A), a rotational rule defined concentric forms (Figure 1B, and a vertical displacment rule defined the translational forms (Figure 1C). For random Glass patterns (Figure 1D), the direction of the displacement was randomly determined for each dipole while the dipole separation distance was kept constant. For our studies, we used vertical translational patterns because other studies have shown that they are more salient than horizontal translational patterns (Appelle, 1972; Jenkins, 1985). For Experiments 1 and 2, the dots were white dots on a midgray background, and the extent of the pattern measured 12° in length and 8° in width. For Experiments 3 and 4, half the dipoles were white and the other half black on a midgray background, and the extent of the pattern measured 21.5° in length and 14.5° in width. The Glass patterns were altered for the masking experiments, because a series of short pilot experiments (data not shown) suggested that the combination of black and white dipoles made it easier to eliminate ceiling effects without degrading the faces further. All Glass patterns were luminance equated to a mean luminance of 22 cd/m². 

Figure 1

 

Radial (A), concentric (B), translational (C), and random (D) Glass patterns used in this study are depicted.

Faces:

The face stimuli for this study came from three sets: NimStim (Tottenham et al., 2009), Karolinska Institute (Lundqvist, Flykt, & Ohman, 1998), and William Kelley's lab (Wig, Miller, Kingstone, & Kelley, 2004). First, all faces were chosen such that the eyes for each face were located 2.5° vertically from the center of the face. For this study, the faces were then cropped to 12° in length and 8° in width to control for head shape. Poirer and Wilson showed that both head shape (Or & Wilson, 2013; Poirier & Wilson, 2010; Wilson & Wilkinson, 2002) and the internal arrangements of features (Or & Wilson, 2013) contribute to face perception. Since we were testing whether the arrangements of face features and concentric Glass pattern perception share a common processing mechanism, we wanted to rule out the possibility that effects of visual adaptation or masking resulted from head shape perception. By cropping these faces, we also removed extraneous features that aid in face perception, such as hair. Future studies including head shape could extend the work presented here. Faces with eyewear or other discerning features (e.g., facial hair, moles, and emotional expressions) were excluded from the final set. The contrast of the face was measured by calculating the root mean square of the luminance for each face (standard deviation of luminance divided by the mean luminance of all pixels in the face; Oruc & Barton, 2010). All faces were scaled to equate the mean luminance for each face to 22 cd/m², and normalized to equate the root mean square for all faces to 1. 

Flowers:

The flowers (Figure 2A) used in Experiment 1 came from the Microsoft Photo Gallery (http://dgl.microsoft.com). The pictures were converted to grayscale and cropped so that the dimensions of each picture matched the face stimuli (12° in length, 8° in height). The mean luminance and contrast of each picture was equated using the same procedure as with the face stimuli. 

Figure 2

 

Examples of flower (A), inverted face (B), and upright face (C) adaptors used in Experiment 1 are shown here. The N-O mask (D) was used as a process-terminating backward mask in Experiment 2.

Staircasing procedure

To equate task difficulty across all participants, a staircasing procedure was used prior to each experiment to determine the salience of the target. For simplicity, target saliency will be referred to target coherence throughout the manuscript. For Glass pattern targets in Experiment 1, target coherence was measured as the proportion of dipoles following the geometric rule and determined for each subject per target type via the staircasing procedure. For target faces in Experiments 2, 3, and 4, target coherence was measured by swapping a percentage of face pixels with pixels from a noise stimulus and determined for each subject per experiment via the staircasing procedure. The noise stimulus was a 12° × 8° ellipse where the intensity of each pixel within the ellipse was randomly selected from a Gaussian distribution of intensity values. The central tendency of this distribution corresponded to a luminance of 22 cd/m², and the range of intensity values was selected to match the contrast of the face. 

Prior to each experiment, each participant performed a two-interval, two-alternative forced choice discrimination task, where subjects were presented with two stimuli and indicated whether the first or the second stimulus matched the predetermined target. We used a two-interval force choice task in order to eliminate potential bias when evaluating subject-specific thresholds. For example, a trial might begin with a face appearing for a short duration followed by a second face. One of these faces was upside-down and the other upright, and the participant must then indicate whether the first or second face was upright. If the participant correctly responded, the target coherence was reduced by 14%. If the participant incorrectly responded, the target coherence was increased by 50%. The staircasing procedure continued until the subject showed 10 reversals, where the target coherence increased after being reduced or vice versa. The mean measured coherence for the reversals was used as the target coherence in the actual experiment. This procedure has been shown to approximate 78% accuracy (∼1.7 d′ units) for a two-alternative forced choice discrimination task (Garcia-Perez, 1998). 

Analysis

All experiments performed were one-interval, two-alternative forced choice tasks, where participants indicated whether a given target belonged to one of two groups of targets: Group A or Group B. Discriminability was measured by d′, which is derived from signal detection theory (Green & Swets, 1966). This measure can be calculated with the following equation: d′ = Z(Group A targets called A/Group A targets) – Z(Group B targets called A / Group B targets), where Z(p) is the inverse of the cumulative Gaussian distribution. Bias is measured using β. β = Y(Group A targets called A / Group A targets) / Y(Group B targets called Group A / Group B targets), where Y(p) is the probably density function of the normal distribution. In Experiment 1, pattern trials were labeled as Group A, and noise trials were labeled as Group B. In Experiments 2 and 3, upright faces were labeled as Group A, and inverted faces were labeled as Group B. For Experiment 4, male faces were labeled as Group A, and female faces were labeled as Group B. 

Experiment 1: Face adaptation

Methods

Thirty adults (19 male, 11 female; aged 25 ± 2.7 years) participated in this experiment. A schematic of the experiment is shown in Figure 3. Each run comprised 30 trials; 15 trials had random Glass pattern targets, termed noise, and the other 15 had pattern targets. The first trial began with an initial adaptation period of 20 s (Clifford & Weston, 2005). During the adaptation period, a different stimulus of the same category was presented 1 s for every second. After the adaptation period, a target would appear for 492 ms. After hearing a tone, the participant would then respond whether the target was a pattern or noise. Every subsequent trial began with a similar 5-s adaptation period followed by the presentation of a target for 492 ms. There were 12 runs, and the type of adaptor and type of target varied from run to run. The order of the runs was pseudorandomly counterbalanced for the participants. Adaptors were random Glass patterns, flowers (Figure 2A), inverted faces (Figure 2B), and upright faces (Figure 2C). Targets were concentric, radial, or translational Glass patterns. For each subject, target coherence was measured via the staircasing procedure prior to the actual experiment. The timing for the staircasing procedure was the same as the other trials. Each trial started with a 5-s interval where no adaptor was shown. After the 5-s interval, the first Glass pattern was shown for 492 ms. One second after the offset of the first Glass pattern, another Glass pattern was shown for 492 milliseconds. One of the Glass patterns was always random and the other contained a coherent pattern. The subject would indicate whether the former or latter Glass pattern contained a pattern. 

Figure 3

 

Schematic of Experiment 1 paradigm. An explanation of the schematic is provided in the text.

Measures of d′, and β were calculated for each participant. Adaptor by target type (4 × 3) repeated measures ANOVAs were used to test for significant adaptor-target interactions for d′ and β. Significant effects were examined post-hoc using paired t tests. Post-hoc statistics are reported as uncorrected values. 

Results: Adaptation to faces does not impair concentric Glass pattern discrimination

Measures of d′ show that adaptation to flowers impaired radial Glass pattern discrimination, but face adaptation did not impair concentric Glass pattern discrimination (Figure 4). An adaptor by pattern (4 × 3) repeated measures ANOVA revealed a significant adaptor by pattern interaction, F(4.32, 125) = 5.786, p < 0.001, η_p² = 0.166. A flower-specific effect of adaptation was observed for discriminating radial Glass patterns such that flower adaptors reduced d′relative to random, t(29) = 4.4, p < 0.001; inverted face, t(29) = 3.7, p < 0.001; and upright face, t(29) = 4.6, p < 0.001, adaptors. This flower-specific adaptation demonstrates that the timing parameters for the paradigm produced behavioral effects of visual adaptation and that we could generate object-to-Glass pattern cross-adaption. A nonspecific effect of adaptation was observed for discriminating translational Glass patterns such that discriminability was reduced for flower, t(29) = 5.5, p < 0.001; inverted face, t(29) = 5.1, p < 0.001; and upright face, t(29) = 5.3, p < 0.001, adaptors relative to random pattern adaptors. No face-specific effects of adaptation were observed on discriminability of concentric Glass patterns (p > 0.42). 

Figure 4

 

Discriminability is plotted for the 12 runs performed during the experiment. Each column represents runs where the pattern targets were of a particular type (radial, concentric, or translational). The solid line represents the expected performance for the participants in the absence of an adaptor as determined by the staircasing procedure. Each row represents the type of adaptor stimulus used in each block. Flower adaptation impaired radial Glass pattern discrimination (dotted black circle) relative to face and random adaptation (dotted black lines). Flower and face adaptation impaired translational Glass pattern discrimination relative to random adaptation (gray circles). The targets shown in the figure are enlarged to make the global forms visible; the actual size of the targets is described in the General methods section.

Face and flower adaptation affected bias for concentric Glass pattern discriminations, but no other effects were found. Measures of β are shown in Figure 5. An adaptor by pattern (4 × 3) repeated measures ANOVA revealed a significant adaptor by pattern interaction, F(3.53, 174) = 3.1, p = 0.023, η_p² = 0.097. Post-hoc pairwise comparisons show that for runs with concentric and noise targets, participants indicated that the target was noise more than concentric during random pattern adaptation relative to flower, t(29) = 2.9, p = 0.006; inverted face, t(29) = 2.5, p = 0.02; and upright face, t(29) = 2.9, p = 0.007, adaptors. No other effects of bias were found (p > 0.1). 

Figure 5

 

β estimates for Experiment 1 are plotted for the 12 runs in Experiment 1. The organization of the figure is the same as in the Figure 4. The solid line represents the β value where bias is neutral.

Discussion

The data presented here show that adaptation to faces does not affect concentric Glass pattern discrimination. Adaptation to flowers impaired radial Glass pattern perception. It is possible that the flowers were effective adaptors because they possess a radial shape, and therefore generated a percept of the same global form as the radial Glass patterns. The flower-adaptation finding is important because the flower-specific adaptation effect on radial Glass pattern discrimination suggests that the paradigm employed can reveal visual adaptation effects of photographic pictures on dot patterns. 

We also found a nonspecific effect on translational Glass pattern perception. Both flower and face adaptation impaired translational Glass pattern perception. This was an unexpected finding. An electrophysiological study of translational Glass pattern perception suggested that such perception utilizes visual processing mechanisms instantiated in V2 (Smith et al., 2007). It is possible that these objects share a common low-level processing mechanism with translational Glass patterns. For example, it has been shown that vertical information may be represented in the center of the eyes, the bridge of the nose, and the edges of the head (Dakin & Watt, 2009). 

Alternatively, one could interpret our findings on translational Glass pattern perception as suggesting that adaptation to random Glass patterns enhances translational Glass pattern discriminability. Such an enhancement may have resulted from the type of staircase procedure used here. We used a two-interval forced choice staircase in order to eliminate potential bias in determining the coherence thresholds (Dixon & Mood, 1948). Future studies could test whether a one-interval staircasing procedure would produce a similar enhancement effect. 

The lack of face-specific adaptation effects on concentric Glass pattern perception probably does not simply reflect a problem with the experimental approach or design implementation. It is possible that adaptation effects are unidirectional (Ross & Edwin Dickinson, 2007). Therefore, Experiment 2 tested whether adaptation to concentric Glass patterns impairs face discrimination. 

Experiment 2: Glass pattern adaptation

Introduction

Previous studies of concentric Glass pattern adaptation (Clifford & Weston, 2005; Ross & Edwin Dickinson, 2007), suggested that adaptation to concentric Glass patterns impairs concentric Glass pattern discrimination. However, these studies also raised the possibility that concentric Glass pattern adaptation effects may be explained as visual masking effects. As described in the Introduction section, effects of visual adaptation last longer than visual masking. If the target appears immediately after the adaptation period, the observed effects of adaptation could possibly be a form of visual masking. The following experiments manipulated the onset of the target relative to the adaptation period to test whether adaptation to (i.e., long duration between onset of target relative to adaptor) or masking with (short duration between onset of target relative to adaptor) concentric Glass patterns impaired subsequent upright/inverted face discrimination. 

Methods

Five adults (two male, three female) participated in this experiment. A schematic of the experiment is shown in Figure 6. Each run comprised 150 trials; the target for 75 trials was an upright face, and the target for the other 75 was an inverted face. Similar to Experiment 1, the first trial began with a 20-s adaptation period, while the other trials began with a 5-s adaptation period. A target face was presented after the adaptation period for 13.3 ms. At 26.6, 40, 66.6, 80, or 106.6 ms after the onset of the target, a backward mask was presented for 292.6 ms. The participant would then indicate whether the target face was upright or inverted. Five conditions were tested to determine whether adaptation to or masking with concentric Glass patterns impaired discrimination between upright and inverted faces (Table 1). The adaptors were either concentric Glass patterns or none. The onset of the target face occurred either immediately (gap absent) or 492 ms (gap present) after the adaptation period. We used process terminating backward masks in order to make the task harder. In pilot experiments (data not shown), we found that people were at ceiling for the task when no mask was used. One backward mask was an N-O mask (see Figure 2D) that was constructed similar to the N-O mask used in previous studies (Rolls, Aggelopoulos, & Zheng, 2003). The N-O mask (Figure 2D) contains contours that might make it more effective as a backwards mask than noise (Rolls et al., 2003). We used a noise mask in one condition because some of our earlier piloting suggested an effect of adaptation when a backward noise mask was used. This potential adaptation effect occurred when the gap between the adaptor and target was absent, and we wanted to see if this adaption effect would occur when the gap was present. 

Figure 6

 

Schematic for the paradigm used in Experiment 2. Explanation of the schematic is provided in the text.

Table 1

 

Conditions in Experiment 2 are shown here. Notes: Row 1: No adaptor control condition. Row 3: The adaptor and face stimuli used in this condition differed from the stimuli used in Experiment 1. Rows 2, 4, 5: Adaptor and face stimuli used in this condition were the same as in Experiment 1.

Condition	Adaptor	Gap	Backwards mask	Face	Stimulus aperture
No-adaptation	none	N/A	N-O	oval	oval patterns
Gap-absent/N-O-mask	concentric	absent	N-O	oval	oval patterns
Gap-absent/whole-faces	concentric	absent	N-O	full view	square patterns
Gap-present/N-O-mask	concentric	present	N-O	oval	oval patterns
Gap-present/noise-mask	concentric	present	noise	oval	oval patterns

The no-adaptation, gap-absent/N–O-mask, gap-present/N–O-mask, and gap-present/noise-mask conditions utilized the stimuli depicted in the General methods section (see Stimulus design). Briefly, Glass patterns were white dots on a gray background. The faces were derived from three face sets and were all mean luminance and contrast equated. Examples of degraded faces are shown in Figure 7. Target coherence was determined for each subject by the staircasing procedure. The timing for each staircasing trial was similar to the other trials in the no-adaptation condition. Each trial started with a 5-s period where no stimulus was presented. Afterwards, the first face would appear for 13.3 ms. Then, a random Glass pattern was presented 66.6 ms after every target as a backward process-terminating mask. The random Glass pattern mask had a duration of 292.6 ms. One second after the random Glass pattern mask disappeared, another face would appear for 13.3 ms; 66.6 ms after the face appeared, another random Glass pattern mask appeared for 292.6 ms. Subjects indicated whether the first or second face was upright. 

Figure 7

 

Examples of degraded faces are shown here. For Experiments 2 and 3, the median percentage of face pixels swapped for upright (A) and inverted (B) faces was 77%. For Experiment 4, the median percentage of face pixels swapped for male (C) and female (D) faces was 50%.

The gap-absent/whole-faces condition in this study used different stimuli from Experiment 1 to determine whether potential masking effects would be affected by different luminance conditions. Briefly, faces were either upright or inverted full-view faces embedded in snow. The snow was defined by a random Gaussian distribution around a mean pixel intensity of 128. The snow image was then averaged with the face; for each pixel in the target, the pixel intensity for that location was the mean of the face pixel and the snow pixel at that location. The snow was also used as the mask in the gap-present/noise-mask condition. No staircasing procedure was used to determine subject-specific target coherence levels for the gap-absent/whole-faces condition. Glass patterns were constructed using MATLAB and the Psychophysics Toolbox. Patterns were presented as white dots on a midgray background. Dots measured 0.04° (one pixel). Dipole separation was 0.12° Dot density was 25%. The stimulus extent was square to match the stimulus extent of the faces. The mean pixel intensity of the glass patterns was approximated to the mean pixel intensity of the faces. 

Measures of d′ and β were calculated for each participant. For d′ and β, stimulus onset asynchrony (SOA) by condition (5 × 5) repeated measures ANOVAs were used to test for significant interactions. Significant effects of condition for specific SOAs were subsequently tested using paired t tests. Post-hoc statistics are reported as uncorrected values. 

Results: Concentric Glass pattern masking, but not adaptation, impairs upright/inverted face discrimination

The conditions tested show differences in discriminability. Measures of d′ are depicted in Figure 8 (top). A condition by SOA (5 × 5) repeated measures ANOVA shows a significant condition by SOA interaction, F(16, 64) = 2, p = 0.023, η_p² = 0.338. β measures are depicted in Figure 8 (bottom). A condition by SOA (5 × 5) repeated measures ANOVA shows no significant condition by SOA interaction, F(16, 64) = 0.39, p = 0.98, η_p² = 0.089. A main effect of condition was observed, F(4,16) = 3.1, p = 0.044, η_p² = 0.439; however, post-hoc comparisons do not reveal significant differences between any of the conditions at any of the SOA levels, largest t(4) = 2, smallest p = 0.113. 

Figure 8

 

Measures of discriminability (top) and bias (bottom) for the conditions in Experiment 2. The no-adaptation condition (solid black) is compared with the noncontrol conditions (dotted black): (A) gap-absent/N–O-mask condition, (B) gap-absent/whole-faces condition, (C) gap-present/N–O-mask condition, (D) gap-present/noise-mask condition. For clarity, examples of target faces (left) and masks (right) in the adaptation conditions are shown above. These stimuli are not drawn to scale.

Post-hoc comparisons show that discriminability was reduced in conditions where the gap was absent and not when the gap was present. Conditions in which the gap between adaptation and target presentation was absent (i.e., gap-absent/N–O-mask and gap-absent/whole-faces conditions) reduced d′ at the 66.6-ms (gap absent/N-O mask: t[4] = 3.5, p = 0.024; condition gap absent/whole faces: t[4] = 6.8, p = 0.002) SOA relative to the no-adaptation condition. A trend was observed, such that the gap-absent/N–O-mask condition reduced discriminability relative to the no-adaptation condition at the 40-ms SOA, t(4) = 2.7, p = 0.054, while the gap-absent/whole-faces condition reduced discriminability relative to the no-adaptation condition at the 40-, t(4) = 2.9, p = 0.044; 80-, t(4) = 4.7, p = 0.009; and 106.6-ms, t(4) = 7.0, p = 0.002, SOAs. Conditions where the gap was present produced no reduction in discriminability across the SOAs. The gap-present/N–O-mask condition showed no effects at all on discriminability (p > 0.4), while gap-present/noise-mask condition showed increased discriminability for the 40- (t(4) = 3.3, p = 0.031), and 80-ms (t(4) = 3.3, p = 0.031), SOAs. 

Discussion

The gap-absent/N–O-mask and gap-absent/whole-faces conditions showed that adaptation to concentric Glass patterns impaired face perception when the gap between adaptor and target was absent. In the gap-present/N–O-mask and gap-present/noise-mask conditions (the gap between the adaptor and target was present), concentric Glass patterns did not impair face perception. The impairment observed for the gap-absent/whole-faces condition may have resulted from differences in the face stimuli. Because we did not use a staircase for these stimuli, the whole faces may have been harder to discriminate than the oval faces. Alternatively, concentric Glass patterns may have adapted face perception in this condition because of the presence of additional face properties such as head shape. Regardless, an effect of masking was observed for the arrangement of face features in the oval faces, which suggests that concentric Glass patterns mask the arrangements of face features. 

The gap-present/noise-mask condition shows enhanced performance relative to the no-adaptation condition. This enhancement is likely unrelated to the concentric Glass patterns themselves, since the gap-present/N–O-mask condition used the same stimuli but did not show this enhancement effect. The parsimonious explanation is that the noise mask is a less effective mask than the N-O mask. However, it is possible that the noise mask, itself, may have somehow enhanced the ability to perceive the face. Other experiments would be needed to draw a more definitive interpretation. Because this effect seems unrelated to the faces and concentric Glass patterns used, we will not discuss it further, here. 

Although a main effect of condition was observed on β, post-hoc comparisons revealed no significant effects. The standard error (SE) for the no adaptation condition was extremely large, and the large SE might explain why no effect of β was observed in our post-hoc comparisons. Because of the variability in β per condition, it is difficult to draw any conclusions from the main effect of β observed here. Future studies could examine whether the effect of bias observed here is significant. 

Findings from the second experiment suggest that masking, but not adaptation, of faces by concentric Glass pattern, impairs face perception. The second experiment shows that impairment only occurs when the adaptor and target are presented close together. Experiments 3 and 4 were then run to test whether this masking effect is specific to concentric Glass patterns, by testing whether masking of faces using concentric Glass patterns, more so than radial or translational Glass patterns, impairs upright/inverted and/or face gender discriminations. 

Experiment 3: Upright/inverted face discrimination

Methods

Thirty-four adults (15 male, 19 female; aged 23.6 ± 3.1 years) participated in a sandwich-masking experiment, where both a forward and backward mask is presented for each target (e.g., Huang, Xiang, & Cao, 2006). A schematic of this experiment is provided in Figure 9. At the start of each trial, a forward-masking Glass pattern would appear for 40 ms; 492 or 66.6 ms after the onset of the mask, a target face would appear for 13.3 ms. Vertical offset conditions were employed to prevent participants from only using the eyes or mouth to determine whether the face was upright (Figure 10A) or inverted (Figure 10B). For the vertical offset conditions, the face moved and the Glass pattern stayed in the same place. The center of the target face could be located in the center of the screen, shifted vertically up 4.1° from the center, or shifted vertically down 4.8° from the center. These shifts ensured that the eyes for the upright face in the center were located in the same position as the eyes for the inverted face shifted upwards, and that the eyes for the inverted face in the center were located in the same position as the eyes for the upright face shifted downwards. The center of the Glass pattern masks was always located at the center of the screen, and the Glass patterns masks always covered the whole face. The target would appear, then 492 or 66.6 ms would elapse, followed by a backward masking Glass pattern of the same type as the first pattern for 40 ms. The SOA for the forward and backward masks were always the same duration. Unlike adaptation, masks affect target discrimination for a very short time. Therefore, the 492-ms presentation condition is a control condition where no masking is expected. The participant would then indicate whether the target face was upright or inverted. For each trial, both masks were concentric, radial, or translational Glass patterns. Per subject, target coherence was measured prior to the experiment via the staircasing procedure. The timing of the staircasing procedure was similar to the masking experiment. Each staircasing trial started with a 425.6-ms period where no stimulus was shown. A random Glass pattern mask was then presented for 40 ms; 66.6 ms after the mask, a face was shown for 13.3 ms. After the onset of the mask appeared for 40 ms. After the offset of the random Glass pattern mask, 425.6 ms elapsed, then another random Glass pattern mask was shown for 40 ms. After the onset of the next mask, 66.6 ms elapsed, then the second face appeared for 13.3 ms. After the onset of the face, 66.6 ms elapsed, and another random Glass pattern mask appeared. Subjects indicated whether the first or second face was upright. The dotted line in Figures 11 and 12 represents the expected performance for these 66.6-ms SOA random Glass pattern masks. 

Figure 9

 

Schematic for the paradigm used in Experiments 3 and 4.

Figure 10

 

Examples of the relative location of the offset (light) and center (dark) faces for Experiments 3 and 4. The offsets were shifted to provide positional ambiguity for the eyes in the center face. (A) The top face is an inverted face shifted up from the middle; the middle face is an upright face; the bottom face is an inverted face shifted down from the middle. (B) The top face is an upright face shifted up from the middle; the middle face is an inverted face; the bottom face is an upright faces shifted down from the middle.

Figure 11

 

Discriminability (A) and bias measures (C) for Experiment 3 are plotted for targets located in the center. Discriminability (B) and bias measures (D) for targets located offset from the center are also plotted. Lines and bars represent concentric (dotted gray), radial (solid gray), and translational (solid black) maskers. A pattern-specific effect of visual masking was found such that concentric masks impaired discriminability more than radial masks, which impaired discriminability more than translational masks (black ellipsoid). The dotted line represents the expected face discrimination performance with random Glass pattern masks at the 66.6-ms SOA, as determined by the staircasing procedure.

Figure 12

 

Discriminability (A), and bias (B) measures plotted for Experiment 4. Lines and bars represent concentric (dotted gray), radial (solid gray), and translational (solid black) maskers. Pattern-specific masking effects were observed such that concentric masks reduced face discriminability more than radial masks, which reduced face discriminability more than translational masks (black ellipsoid). The dotted line represents the expected gender discrimination performance with random Glass pattern masks at the 66.6-ms SOA (A) and the β value where bias is neutral (B).

Measures of d′, and β were calculated for each participant. A priori, no difference between the down and up offset conditions was predicted, and the down and up offset conditions were collapsed across each other to simplify the analysis. For d′ and β, SOA by offset by pattern (2 × 2 × 3) repeated measures ANOVAs were used to test for significant pattern-SOA interactions. Effects between individual conditions were evaluated with subsequent post-hoc t tests. Post-hoc statistics are reported as uncorrected values. 

Results: Concentric Glass pattern sandwich masking impairs upright/inverted face discrimination

Measures of d′ show a pattern-specific effect of visual masking, such that concentric Glass patterns impaired face discrimination more than radial Glass patterns, and radial Glass patterns impaired face discrimination more than translational Glass patterns (Figure 11). A pattern by SOA by offset repeated measures ANOVA shows a significant pattern by SOA by offset interaction, F(2, 66) = 3.5, p = 0.033, η_p² = 0.098. Post-hoc comparisons show a pattern specific masking effect for central face presentations (Figure 11A), such that concentric Glass pattern masking reduced d′ at the 66.6-ms SOA more than radial, t(33) = 3.0, p = 0.005, or translational, t(33) = 5.7, p < 0.001, Glass patterns. Radial Glass pattern maskers reduced d′ at the 60-ms SOA more than translational patterns, t(33) = 2.2, p = 0.039. When the face was presented offset from the center of the screen (Figure 11B), no significant pattern by SOA interactions were observed, F(2, 66) = 0.431, p = 0.65, η_p² = 0.013. 

There was no pattern-specific effect of masking on bias (Figure 11C, D). A pattern by SOA by offset (3 × 2 × 2) repeated measures ANOVA revealed no significant pattern by SOA interaction, F(1.7, 55.8) = 1.08, p = 0.337, η_p² = 0.032. 

Discussion

The results from Experiment 3 show a pattern-specific masking effect; concentric Glass pattern masking of faces impaired upright/face discrimination more than radial or translational pattern masks. This evidence supports the hypothesis that concentric Glass patterns and faces share a common processor. Furthermore, face discrimination performance during radial Glass pattern masking of faces was similar to the expected face discrimination performance for random Glass pattern masking of faces, suggesting that a shared processor exists for concentric, but not radial, Glass pattern and face perception. Previous visual adaptation and visual masking studies support this distinction. While one visual adaptation study suggested that radial and concentric Glass patterns might share a common processing mechanism (Clifford & Weston, 2005), another visual adaptation study suggested that radial and concentric Glass pattern perception use different mechanisms for processing the global form (Ross & Edwin Dickinson, 2007). Radial Glass pattern masking of concentric Glass patterns does not affect concentric Glass pattern perception (Chen, 2009). One fMRI study showed that concentric, but not radial, form presentation increases activation of visual regions involved in face processing (Wilkinson et al., 2000), suggesting that face perception may utilize a processing mechanism shared by concentric, but not radial, Glass pattern perception. Another fMRI study demonstrated that blood-oxygen-level-dependent (BOLD) responses become increasingly selective for concentric Glass patterns for successively higher-level visual areas (Ostwald et al., 2008), suggesting that the high-level visual processing of radial and concentric Glass patterns may differ. 

Radial Glass patterns impaired upright/face discrimination more than translational patterns, suggesting either that translational Glass pattern masking of faces enhanced face perception or that the effects of Glass pattern masking of faces are not uniform across the three types of Glass patterns (see the General discussion section for a discussion of these alternative explanations). 

These effects were attenuated when the faces were offset from the center of the Glass pattern. Discriminability of the offset targets does not appear to be different from discriminability of the center targets. Therefore the attenuation of the masking effects was not driven by an increased difficulty in discrimination of peripheral targets. The attenuated effects on offset targets suggest three points. 

Because the offsets created positional ambiguity for the location of face features, participants may have been unable to utilize a feature-based strategy (that they might have adopted at the offset locations) for performing the task when the face was centered. The offset conditions, therefore, forced the participants to make judgments based on the arrangements of the features (i.e., whether the eyes are above or below the mouth) when the face was centered. Because this ambiguity was not present for the offset conditions themselves, a feature-based strategy may have been used for the offset conditions. 

However, the lack of any masking effect at the offset positions may suggest that the processing of arrangements of features involves more than simply processing the geometric rule, and that such processing may only affect face perception when the face is located in the center of the Glass pattern. This suggests that the processing of arrangements of face features may be important for determining face symmetry (Poirier & Wilson, 2010), and that concentric Glass pattern perception may also utilize the same processing mechanism. Such processing may reflect the pooling of contour information of the internal face features or the pooling of contour information of the concentric or radial Glass patterns (Poirier & Wilson, 2010). 

Finally, the lack of masking effects at the offset positions suggests that the shared mechanism may be instantiated at a visual processing level such as V4. Neurons in face-specific regions are thought to have large receptive fields that encode for whole hemifields (Boussaoud, Desimone, & Ungerleider, 1991; Chelazzi, Duncan, Miller, & Desimone, 1998; Desimone & Gross, 1979; Niemeier, Goltz, Kuchinad, Tweed, & Vilis, 2005; Op De Beeck & Vogels, 2000). Shifting the face away from the center of the Glass pattern attenuates the masking effect, suggesting that the neurons affected by masking may have smaller receptive fields. Although speculative, the idea that V4 instantiates this shared mechanism is consistent with predictions from a model of symmetry perception that is tuned to both concentric Glass patterns and faces (Poirier & Wilson, 2010). Future studies could manipulate the location of the center of the masking Glass pattern with respect to the face target to see where the pattern-specific masking effects are the strongest. 

A previous fMRI study has suggested that concentric Glass patterns and face perception share a common processing mechanism instantiated by face-specific visual cortical regions (Wilkinson et al., 2000). Although the masking effect observed here does not implicate face-specific regions, it is certainly possible that functional overlap is present in both V4 and face-specific regions. Since we did not examine effects of head contour here, perhaps face-specific regions are involved in integrating head contour and internal feature arrangement information. Future studies could also test whether head contour and internal feature arrangements may be processed synergistically. Recent work has shown that face perception involves processing the head shape and the internal feature arrangements (Or & Wilson, 2013), and it is possible that the combination of the two processes is important for face perception (Poirier & Wilson, 2010). Such studies could examine potential interactions between concentric Glass patterns, head contours, and internal face features. 

Experiment 4 extends our finding of form-specific Glass pattern effects of masking on upright/inverted face judgments by testing whether concentric, more so than radial and translational, Glass pattern masking of faces impairs face gender discrimination. Unlike inversion, gender is a high-level property of a face. Therefore, a finding that concentric Glass pattern masking of faces impairs gender discrimination would provide convergent evidence that concentric Glass patterns and faces share a common processing mechanism. 

Experiment 4: Gender discrimination

Methods

Thirty adults (12 male, 18 female; aged 23.4 ± 3.2 years) participated in this experiment; these adults also participated in Experiment 3. Experiment 4 was similar to Experiment 3 (Figure 9); the target faces in Experiment 4 were male (Figure 7C) and female (Figure 7D) faces, and no vertical offset conditions were used. Per subject, target coherence was measured prior to the experiment via the staircasing procedure. The same procedure was used as in Experiment 3, except that subjects indicated whether the first or second face was male instead of upright. The dotted line represents the expected performance, as determined by staircasing, for random Glass pattern masks at the 66.6-ms SOA. 

Measures of d′ and β were calculated for each participant. For d′ and β, SOA by pattern (2 × 3) repeated measures ANOVAs were used to test for significant pattern SOA interactions. Effects between individual conditions were evaluated with subsequent post-hoc t tests. Post-hoc statistics are reported as uncorrected values. 

Results: Concentric Glass pattern sandwich masking impairs gender discrimination

Consistent with the results from Experiment 3, concentric Glass pattern masks reduced d′ relative to radial Glass patterns, which reduced d′ relative to translational masks (Figure 12A). A pattern by SOA (3 × 2) repeated measures ANOVA revealed a significant pattern by SOA interaction, F(2, 58) = 4.5, p = 0.015, η_p² = 0.134. Post-hoc comparisons show pattern-specific effects of masking at the 66.6-ms SOA, such that concentric Glass pattern masking reduced d′ relative to radial, t(29) = 3.0, p = 0.005, and translational, t(29) = 4.7, p < 0.001, masking. Radial Glass pattern masking reduced d′ relative to translational masking, t(29) = 2.6, p = 0.015. 

No effects of masking on bias were observed (Figure 12B). A pattern by SOA (3 × 2) repeated measures ANOVA shows no significant pattern by SOA interaction, F(1.67, 48.3) = 0.572, p = 0.538, η_p² = 0.019. 

Discussion

The results from Experiment 4 provide convergent evidence for a pattern-specific masking effect on face discrimination. Since two different face discrimination tasks reveal pattern-specific masking effects, it is less likely that the pattern-specific masking effect on faces interrupted a low-level processing mechanism. Therefore, the evidence from Experiment 4 suggests an interaction between concentric forms and face perception affected processing in mid- or high-level visual regions. As with Experiment 3, both concentric and radial masking of faces impaired performance more than translational Glass pattern masking of faces, which may suggest a processing interaction between concentric and radial forms and face perception, or that translational Glass pattern masking of faces enhanced face perception (see the General discussion section for a discussion of these alternative explanations). 

General discussion

Faces are complex stimuli. Despite their complexity, humans are able to perceive and discriminate faces rapidly (Lehky, 2000; Tsao & Livingstone, 2008). Studies have shown that face perception may involve processing the arrangements of the face features (Boutsen et al., 2006; Lewis & Johnston, 1997; Maurer et al., 2002; Mondloch & Maurer, 2008; Rhodes et al., 1993; Tanaka & Farah, 1993; Thompson, 1980; Young et al., 1987), and others hint at a possible interaction between concentric Glass patterns and faces (Aspell et al., 2006; Ostwald et al., 2008; Poirier & Wilson, 2010; Rentschler et al., 1994; Smith et al., 2007; Wilkinson et al., 2000; Wilson & Wilkinson, 1998; Wilson et al., 1997). The observation that concentric Glass pattern masking of faces impaired performance more than translational Glass pattern masking of faces suggests that concentric Glass pattern perception is more similar to face perception than translational Glass pattern perception. A parsimonious explanation for this perceptual similarity is that face perception utilizes a processing mechanism shared by concentric Glass pattern perception. 

Differences between concentric, radial, and translational Glass pattern masking effects on face discrimination

As discussed in Experiment 3, the parsimonious explanation of our results is that concentric, but not radial or translational, Glass patterns and faces share a common processing mechanism. However, we also found that radial Glass pattern masking of faces impaired face perception more than translational Glass pattern masking of faces. One possible explanation is that translational Glass pattern masking of faces enhances face perception. Some have argued that the eyes, nose, and edges of the head contain vertical information, which is represented in translational Glass patterns. If so, this aspect of face processing may share a common processing mechanism with translational Glass patterns (Dakin & Watt, 2009). 

Alternatively, faces and both radial and concentric Glass patterns share a common processing mechanism. The notion of a shared processor for face, concentric, and radial Glass pattern perception is consistent with findings from previous studies. Psychophysical data have shown that processing the arrangements of features is important for concentric and radial, but not translational, Glass patterns. It is easier to detect concentric and radial, but not translational, Glass patterns when the whole pattern is presented (Wilson & Wilkinson, 1998). While larger concentric Glass patterns are more quickly discriminated, larger translational Glass patterns are more slowly discriminated (Aspell et al., 2006). The masking study mentioned in the discussion of Experiment 3 showed that translational maskers of concentric or radial Glass patterns do not affect concentric or radial Glass pattern discrimination (Chen, 2009). The fMRI study, mentioned in the discussion of Experiment 3, also showed that regions sensitive to objects responded to concentric and radial, but not translational forms (Wilkinson et al., 2000), while another revealed that concentric and radial Glass patterns show increased responses for midlevel visual regions (Mannion et al., 2013). 

While the literature suggests that translational Glass patterns utilize different mechanisms compared to radial and concentric Glass patterns, our study does not provide further evidence for whether there are differences in visual processing mechanisms for different Glass pattern types. It is possible that the different types of Glass patterns utilize the same processing mechanisms but that the way these mechanisms are utilized is not the same between the Glass pattern types. 

Effects of visual masking suggest that concentric Glass pattern and face perception share a common processing mechanism

A biologically plausible model of shape perception finds symmetry in both faces and concentric Glass patterns (Poirier & Wilson, 2010). The model uses the shape of the head to determine the center of the face. However, the authors describe a generalized model that might use the arrangements of the face features to identify symmetry. This generalized model suggests that the arrangements of face features and concentric Glass patterns use a common mechanism, and this hypothesis is supported by our pattern-specific masking effects. Concentric Glass pattern masking of faces impairs face perception when the arrangements of the face features are present and head shape is controlled. 

Briefly, the Poirier and Wilson model is divided into five stages. (1) The contour information of an object is recovered using filters that encode contours. (2) Coarse, large-scale, filters that encode the center of concentric contours are used to recover the center of the object. (3) Multiple oriented filters were used to recover local curvature information of contours relative to the center of the object. (4) The information from the first few stages is pooled to determine the shape of the object based on curvature strength. (5) This information is used to identify the axes of symmetry within an object. Each of these processes is thought to be instantiated by successively higher levels of visual processing. A shared processor could involve any of these stages of processing. Future studies may use fMRI to identify putative regions that represent such a shared processor. The location of such regions may suggest which stages of the model may represent the shared process and provide insight into processing mechanisms involved in face perception. Based on our data, we predict that the shared mechanism likely reflects third or fourth stages because concentric pattern masking impaired face discriminations more than radial pattern masking. Since concentric and radial Glass patterns have different local curvature information, these findings suggest that the masking effect occurs after the second stage. As discussed in Experiment 3, it is possible that concentric and radial Glass pattern masking of faces may affect neurons within V4, because the masking effect varies depending upon the position of the face with respect to the Glass pattern. 

The different findings from the adaptation and masking experiments suggests a potential neurophysiological explanation

The positive finding in the visual masking of faces experiments, when contrasted with the negative finding for the visual adaptation of faces experiments, reveals a behavioral phenomenon that has a potential neurophysiological interpretation. Though one should regard this interpretation as somewhat speculative, it may provide a direction for future studies, and is therefore important to include here. 

One explanation for these findings is that the processing mechanism affected by visual masking of faces may be robust to adaptation itself. Neurons instantiating this mechanism may not habituate for long periods of time. Studies have shown altered neuronal tuning curves following visual adaptation (Carandini, Movshon, & Ferster, 1998; Dragoi, Sharma, & Sur, 2000). This alteration may occur because these neurons habituate to the presentation of the adaptor. A review of visual masking studies suggests that masking suppresses neuronal activity, but does not cause the neurons to habituate for long periods of time (Breitmeyer, 2007). 

Interestingly, previous studies of Glass pattern adaptation of Glass patterns have not examined the duration of the adaptation effects on Glass pattern perception; the target was presented immediately after the adaptors (Clifford & Weston, 2005; Ross & Edwin Dickinson, 2007; Vreven & Berge, 2007). It is possible that these adaptation effects are actually effects of visual masking of Glass patterns, and therefore this processing mechanism may not be affected strongly by visual adaptation. Future adaptation studies of Glass pattern adaptation effects on Glass pattern perception can test this by modifying the onset of the target relative to the adaptation period. If a 500-ms gap between the target and adaptation period does not impair target discrimination, then the processing mechanism involved in perceiving concentric Glass patterns may not be affected by visual adaptation. 

Alternatively, it is possible that our lack of adaptation effects stems from our use of cropped faces. By cropping the faces, we eliminated head shape as a source of information. Although we found evidence that concentric Glass pattern masking of faces impaired the processing of the arrangements of face features, it is possible that the internal face features and head shape are processed synergistically (Or & Wilson, 2013; Poirier & Wilson, 2010). If so, it is possible that effects of concentric Glass pattern adaptation on face perception may only be observed when both the head contours and the internal face features are presented. Future experiments could manipulate the presence or absence of internal face features and head shape to test whether concentric Glass pattern adaptation of faces impairs face perception in the context of head shape. 

Our findings are not inconsistent with Loffler et al. (2005), who found that backward masking of faces impaired face perception more than forward masking of faces. Except in Experiment 1, we used a backward mask to increase the difficulty of the task. Without this backward mask, subjects might have been at ceiling and we might not have been able to examine the effects of pattern-specific masks on face perception. In fact, our critical Experiments 3 and 4 used both forward and backward masks to test for pattern-specific masking effects. Taken together, these data suggest that face perception is an extremely difficult process to override. 

Interestingly, Experiment 1 revealed no effects of face masking of concentric Glass patterns on concentric Glass pattern perception. It is possible that the masking effects observed in our study are uni-directional. A previous visual adaptation study showed that visual adaptation effects on static and dynamic Glass patterns are uni-directional (Ross & Edwin Dickinson, 2007). However, it is important to note that targets were not backwards nor sandwich masked in Experiment 1. As Loffler et al. (2005) showed, backwards masking of faces is more effective than forward masking of faces. Therefore, such a difference in our masking experiments may explain why concentric Glass pattern masking of faces impaired face perception, but face masking of concentric Glass patterns did not impair concentric Glass pattern perception. 

Implications for object recognition

The findings here do not test whether Glass pattern masking of nonface objects impairs nonface object discriminations, so there are several possible implications for the nature of the shared processor. One possibility is that faces, and not any other objects, engage this processor. A number of fMRI studies have shown that faces activate some visual regions that are not activated by nonface objects (for a review see Tsao & Livingstone, 2008), and some argue that this difference is reflected in a processing mechanism dedicated only to face perception (Kanwisher, 2000, 2010). 

Alternatively, the shared processor may respond to both face and expertly perceived nonface objects (e.g., dogs perceived by dog experts). Processing feature arrangements may be important for discriminating both faces and expertly perceived nonface objects (Diamond & Carey, 1986; Gauthier & Tarr, 1997). Processing expertly perceived objects may involve some neural regions thought to be specific to face perception (Gauthier, Tarr, Anderson, Skudlarski, & Gore, 1999; Tarr & Gauthier, 2000). 

It is also possible that general form perception, and therefore all object (expert or nonexpert) perception, utilize this shared processor. Visual area V4, which is involved in form perception (Hegde & Van Essen, 2007; Kobatake & Tanaka, 1994), may instantiate part of such a mechanism. Psychophysical (Aspell et al., 2006), computational (Poirier & Wilson, 2010), and neurophysiological (Ostwald et al., 2008) studies suggest that V4 may be necessary for concentric Glass pattern perception. A case study of a lesion in putative V4 showed disrupted form perception, and the participant could not discriminate concentric Glass patterns (Gallant, Shoup, & Mazer, 2000). 

Future studies may investigate whether concentric or radial masking of nonface objects is observed for expertly (e.g., dogs perceived by dog experts) and nonexpertly (e.g., houses) perceived stimuli. Such evidence may help characterize the nature of the shared processor. If concentric/radial masking of expertly perceived objects, more so than translational masking of expertly perceived objects, impairs perception of expertly perceived objects, then expertly perceived objects may engage the shared processor. The shared processor may be involved in general form perception if concentric/radial masking of any nonface object impairs the perception of any type of object more so than translational masking. 

Implications for clinical disorders

The possibility of a shared mechanism between face, concentric, and radial Glass pattern perception may provide insights into clinical disorders involving face-processing deficits, such as autism or prosopagnosia. Both acquired visual agnosia (Gallant et al., 2000) and acquired prosopagnosia (Rentschler et al., 1994) may impair concentric Glass pattern perception. People with autism have face perceptual deficits (Gauthier, Klaiman, & Schultz, 2009; Hadjikhani et al., 2004; Joseph & Tanaka, 2003; Klin & Jones, 2008; Pellicano, Jeffery, Burr, & Rhodes, 2007; Rutherford, Clements, & Sekuler, 2007; Scherf, Behrmann, Minshew, & Luna, 2008; Spezio, Adolphs, Hurley, & Piven, 2007; Teunisse & de Gelder, 2003; van Kooten et al., 2008) that may relate to a general deficit in processing the arrangements of features (Behrmann, Thomas, & Humphreys, 2006). Psychophysical studies of concentric and radial Glass pattern perception in people with autism or prosopagnosia may provide insight into the nature of these processing deficits. 

Conclusions

The combined results of our four experiments provide evidence of a common processor for the arrangements of face features and concentric form-from-structure stimuli. The absence of an adaptation effect suggests that this mechanism may not easily habituate. Perhaps combining face feature arrangements with head shape outlines may be necessary to test potential adaptation-based interactions, in addition to the masking effects we have shown. These findings are consistent with a literature that has hinted at a shared face/concentric processor. However, this processor may also be involved in other types of object perception. The difference between concentric, radial, and translational Glass pattern masking of faces suggests that different types of Glass pattern perception may utilize different types of processing mechanisms or utilize the same processing mechanisms differently. Future behavioral and neurophysiological visual masking studies may identify the location and characteristics of the affected processing mechanism, and whether this mechanism may be compromised in disorders involving face perception. 

Acknowledgments

This work was supported by the McDonnell Center for Systems Neuroscience. Eric Feczko's effort was supported by T32 EY013360. John Pruett's effort was supported by K12 EY016336. We thank Sridhar Kandala for coordinating recruitment, scheduling, and assessments of the subjects. We would like to thank Gagan Wig and Maital Neta for contributing face stimulus sets. We would like to thank Bradley Schlaggar and Larry Snyder for providing advice on experimental design. We would like to thank Richard Mulligan and Sridhar Kandala for helping to proofread the manuscript. We would like to thank the Washington University School of Medicine Volunteers for Health for providing sources for recruitment. 

Commercial relationships: none. 

Corresponding author: Eric Feczko. 

Email: [e-mail]. 

Address: Department of Developmental and Cognitive Neuroscience, Emory University, Atlanta, GA, 30329, USA.