Faces in commonly experienced configurations enter awareness faster due to their curvature relative to fixation

Introduction

The extent to which invisible stimuli are still processed has become a popular line of research over the last decades (Dehaene & Changeux, 2011; Hesselmann & Moors, 2015). One particularly compelling paradigm to render visual stimuli invisible is continuous flash suppression (CFS) (Tsuchiya & Koch, 2005). In CFS, a salient dynamic pattern composed of various colored shapes is presented to one eye while another stimulus is presented to the other eye. Due to the dynamic nature of the mask, the other stimulus is perceptually suppressed and invisible to observers for a time period on the order of seconds. CFS has been implemented in various ways to study processing of perceptually suppressed stimuli, one being the breaking CFS paradigm (b-CFS) (Stein, Hebart & Sterzer, 2011; Gayet, Van Der Stigchel & Paffen, 2014). Here, the contrast of the initially suppressed stimulus is gradually increased until it causes a perceptual breakthrough (i.e., becomes detectable to the observer). The breakthrough or suppression time is then used as an index of the strength of the representation of that visual stimulus during suppression. That is, as in regular binocular rivalry, ‘stimulus strength’ is predicted to influence suppression durations such that stronger stimulus representations break CFS faster than weaker stimuli (Jiang, Costello & He, 2007; Stein, Hebart & Sterzer, 2011). Here, it should be noted however that ‘stimulus strength’ is not a well-defined construct and that there is some debate as to which factors contribute exactly to differences in suppression times. That is, the factors driving breakthroughs could be low-level or high-level (Gayet, Van Der Stigchel & Paffen, 2014; Hesselmann & Moors, 2015), or involve the feature overlap between the CFS mask and the suppressed stimulus (Yang & Blake, 2012; Moors, Wagemans & de-Wit, 2014).

A number of studies have considered the degree to which face stimuli are still processed while perceptually suppressed and have used the b-CFS paradigm, amongst others, to tackle this question (for a review of unconscious face processing, not limited to CFS studies only, see Axelrod, Bar & Rees, 2015). A now-classic study by Jiang, Costello and He (2007) showed that upright face stimuli broke suppression faster than inverted face stimuli, resembling the well-known face inversion effect for consciously presented stimuli (Yin, 1969; Farah, Tanaka & Drain, 1995). Following this study, several b-CFS studies have replicated this face inversion effect (Zhou et al., 2010; Stein, Hebart & Sterzer, 2011; Stein, Peelen & Sterzer, 2011; Stein & Sterzer, 2012; Stein, Sterzer & Peelen, 2012; Gobbini et al., 2013a; Gobbini et al., 2013b; Heyman & Moors, 2014; Stein, End & Sterzer, 2014). Other studies have furthermore indicated that stimulus-related factors such as eye gaze (Stein et al., 2011; Xu, Zhang & Geng, 2011; Chen & Yeh, 2012; Gobbini et al., 2013b), facial expression (Yang, Zald & Blake, 2007; Sterzer et al., 2011; Stein & Sterzer, 2012; Capitão et al., 2014), face identity (Geng et al., 2012; Gobbini et al., 2013a), face race (Stein, End & Sterzer, 2014), or the trustworthiness or dominance of a face (Stewart et al., 2012) can influence suppression times. Taken together, these findings seem to suggest that, while perceptually suppressed, the representation of a face stimulus is a fairly integrated one involving the high-level analysis of several complex features.

In apparent contrast with these b-CFS findings, a more complicated pattern of results has arisen from studies that rely on adaptation to invisible face stimuli or investigate the representation of invisible face stimuli using neuroimaging techniques. For example, adaptation studies have indicated that visual awareness of a face is required for adaptation to complex features such as facial expression (Yang, Hong & Blake, 2010), face race or gender (Amihai, Deouell & Bentin, 2011), face identity (Moradi, Koch & Shimojo, 2005), face shape (Stein & Sterzer, 2011), or eye gaze (Stein, Peelen & Sterzer, 2012). The main conclusion of these studies is that adaptation effects for invisible stimuli are sometimes observed, but they are largely specific to the adapted eye and size of the stimulus. For example, Stein and Sterzer (2011) observed face shape aftereffects for fully invisible stimuli, yet these aftereffects were only observed if the test stimulus had the same size as the adaptor and was also presented to the same eye as the adaptor. This suggests that the adaptation occurred at a low level of processing, and was specific to simple features such as its exact size and shape. Similarly, neuroimaging studies have shown that neural responses to invisible face stimuli are strongly reduced in the fusiform face area (Jiang & He, 2006; Sterzer et al., 2014), although the pattern of activation still enables the successful decoding of certain stimulus distinctions (Sterzer, Haynes & Rees, 2008; Sterzer, Jalkanen & Rees, 2009).

Taken together, behavioral studies relying on adaptation and neuroimaging studies call into question whether the results obtained using the b-CFS paradigm are genuinely attributable to high-level configural processing of the invisible face. Rather, they suggest that the representation of the perceptually suppressed face is limited to lower-level aspects such as its general shape. Therefore, in this study, we were interested to further study the representation of a perceptually suppressed face in a b-CFS context, capitalizing on the findings of a recent neuroimaging study. That is, Chan et al. (2010) recently showed that representations of body parts and faces were strongest in the extrastriate body area and fusiform face area, respectively, when they were presented in their commonly experienced configuration (e.g., the left side of a face presented in the right visual field). This result is intriguing since all conditions simply involved presenting the same stimulus (e.g., right or left side of a face) to a different side of the visual field. Thus, if stimulus strength influences suppression time, we would predict that perceptually suppressed face stimuli presented in their commonly experienced configuration would break suppression faster compared to those presented in the other part of the visual field. Moreover, given that the effect for the face stimuli seems to be specific to the fusiform face area, the presence of such an effect in a b-CFS setup could be indicative of the extent to which invisible face stimuli are processed during suppression. To this end, we also included a face inversion condition. That is, if a congruency effect is observed, this inversion condition will enable us to test whether this effect is dependent on specialized processing for upright faces.

Methods

Stimuli

The background of the display consisted of a random checkerboard pattern to achieve stable binocular fusion. The size of the individual elements of the checkerboard was equal to 0.34°. In both eyes, a black frame (10° by 10°) was superimposed on the checkerboard pattern, onto which the stimuli would be presented. A black (eye dominance measurement) or white (main experiment) fixation cross was continuously present during the experiment (size 0.5 by 0.5°). In the eye dominance measurement phase, the target consisted of an arrow (maximal width 4°, maximal height 2°) and the CFS mask consisted of 150 squares with randomly picked sizes between 1 and 2° and a random luminance value (range: 1–100 cd/m²).

We obtained the stimuli used in the original study of Chan et al. (2010) and used a subset of those in this study (see Fig. 1A). That is, we only used the face configurations of their stimulus set, which consisted of four different half-face exemplars (size 3° of visual angle). For the specific details of the stimulus generation procedure, we refer to the original study. In the main experiment, the CFS mask (6° × 6°) consisted of 200 grayscale squares with a random size between 0.75° and 1.5°. In all parts of the experiments, the CFS mask refreshed its contents every 100 ms (i.e., at 10Hz).

Results

Before subjecting the data to any analysis, suppression times were first log transformed to account for their positive skew. Only correct responses were considered. Outliers were defined as suppression times that deviated more than three standard deviations from the mean suppression time (for each observer separately) and these were also excluded from the analysis. This led to a removal of 5.5% of the data. To facilitate the interpretation of the data, we converted the factors visual field and stimulus side to a single variable termed ‘congruency.’ A congruent stimulus would be one that constitutes a commonly experienced configuration (e.g., right side of the face in the left visual field, assuming fixation in the center). For inverted stimuli, we applied the same transformation such that congruent stimuli would be the ones for which the overall configuration would be the same (e.g., an inverted left side of the face would now have to be presented in the left side of the visual field to be coded as congruent). Table 1 depicts the results of the Bayes Factor analysis. The betting fitting model (BF = 1) is one that includes a main effect of congruency and a main effect of inversion. This model is preferred 5.2 times over a model including also the interaction between the main effects. Furthermore, a model including only a main effect of inversion and no congruency effect is 3.6 times less likely than the best fitting model. For all other models (e.g., a model with a main effect of congruency only), the best fitting model was more than 100 more likely (i.e., BFs > 100). The mean suppression times for all combinations of congruency and face inversion are depicted in Fig. 2. In line with the Bayes Factor analysis, inverted faces yielded longer suppression times than upright faces (the well-known face inversion effect). Furthermore, face stimuli presented in congruent configurations broke suppression faster than the incongruent ones, yet this main effect did was not modulated by stimulus inversion.

Discussion and Conclusion

The goal of this study was to assess whether face stimuli presented in their commonly experienced configurations would break suppression faster than the same stimuli presented in other configurations. Our results indicated that this indeed was the case, yet the effect was not specific for upright face stimuli. That is, similar configurations also broke suppression faster when they were presented inverted rather than upright. This result implies that shape differences relative to fixation were responsible for the observed congruency effect rather than processing mechanisms specific for upright faces.

This study was motivated by the fact that a lot of b-CFS studies on face processing obtained evidence for relatively complex (high-level) processing of invisible faces during CFS. In contrast, studies relying on adaptation or neuroimaging techniques consistently showed that processing of invisible faces is severely reduced compared to visible faces and is possibly only specific to the general face shape rather than the identity, facial expression, or other high-level face attributes. Therefore, we decided to capitalize on the findings of a neuroimaging study in which it was shown that the pattern of responses in the fusiform face area was strongest for face stimuli presented in their commonly experienced (congruent) configuration. Assuming that stimuli with a strong representation break suppression faster, one would predict the same difference between congruent and incongruent configurations to be observed in a b-CFS setup. Moreover, given the specificity of the effect to the fusiform face area, we also predicted that the effect should be absent or at least greatly reduced for inverted faces (Yovel & Kanwisher, 2005). As highlighted above, our results indicated both an effect of configuration as well as inversion but no interaction between those factors. This indicates that the differences in suppression time between conditions are more likely attributable to shape-specific differences between conditions rather than mechanisms relying on the configural processing of faces, which are known to be affected by inversion (Yin, 1969; Farah, Tanaka, & Drain, 1995). Indeed, studies on holistic face perception have shown that face inversion is a stimulus manipulation that strongly influences performance on a wide range of tasks (for a review, see Rossion, 2008; Van Belle et al., 2010). Nevertheless, it has also been argued that inverted faces can still be processed holistically (Richler et al., 2011). This last study has mainly indicated qualitatively similar patterns for upright and inverted faces, but still observed quantitative differences. However, in our study the congruency effect was also quantitatively similar between upright and inverted faces, given the absence of an interaction between face inversion and congruency. Therefore, we think the most parsimonious explanation of our results is one that does not rely on face-specific (high-level) configural processing of perceptually suppressed face stimuli.

One particularly important difference between the stimuli presented in both types of configurations is the curvature of the face shape relative to fixation. That is, in congruent configurations, the curved contour is convex relative to fixation compared to being concave in the incongruent configurations. Several behavioral studies have shown that convex features are often perceptually dominant, for instance, in determining figure-ground relationships or shape similarity (Kanizsa & Gerbino, 1976; Bertamini & Wagemans, 2013). Moreover, neurophysiological recordings have shown a similar bias towards convex features in macaque area V4 (Pasupathy & Connor, 1999). Last, a recent fMRI study has shown that cortical area LOC shows higher sensitivity for convex rather than concave shapes (Haushofer et al., 2008). Although our study only consisted of face stimuli, the pattern of results observed in this study is similar to what would be predicted based on a convexity/concavity account. Thus, in the light of these studies, we can speculate that our findings can be interpreted as potentially reflecting the heightened sensitivity of the visual system to convex features (relative to fixation).

This interpretation is in accord with a larger set of studies that has questioned evidence of high-level processing of stimuli suppressed through CFS. For example, Hedger, Adams and Garner (2015a) recently showed that the advantage of fearful faces breaking suppression faster than neutral ones is predicted by effective contrast of the stimuli. Furthermore, another recent study by the same group observed that attentional orienting due to threat stimuli is completely absent when threatening stimuli were rendered completely invisible (Hedger, Adams & Garner, 2015b). Other studies have cast doubt on whether invisible words can be processed (Heyman & Moors, 2014), numerosity can be extracted during suppression (Liu et al., 2013; Hesselmann et al., 2014; Hesselmann & Knops, 2014), or integration between a suppressed visual looming stimulus and a supraliminal auditory stimulus can occur (Moors et al., 2015).

In sum, the results of this study provide evidence that stimuli that are more strongly represented in the visual cortex break suppression faster than other stimuli. However, the fact that the observed congruency effect was not specific for upright face stimuli indicates that the face stimuli used in this study were presumably not processed by specialized face recognition mechanisms, but rather at a more basic level limited to more general shape properties such as convexity.

Supplemental Information