Francesca van Baarzel
Universiteit van Amsterdam Student number: 10985611 Supervisor: T. Stein Amount of words: 4087
2 Abstract
Faces are harder to recognize when they are presented inverted compared to an upright presentation. A method to further study this effect at an unconscious level is done by a variation of binocular rivalry: continuous flash suppression (CFS). A flashing high-contrast pattern is presented in one eye, while another stimulus of interest is presented in the other eye. Under CFS, this inversion effect causes upright faces to gain privileged access to awareness compared to inverted faces. Here we introduced breaking continuous flash suppression (b-CFS) as a tool to render faces invisible, in order to test if invisible faces would still be
detected above chance level, reflecting blindsight-like effects. In addition, it was examined if face inversion effects would be found for blindsight-like trials. Finally, we tested if the
visibility of faces under b-CFS is a gradual or all-or-none phenomenon by adding a subjective measurement. Subjects had to indicate the rate of visibility on a continuous scale. Results showed blindsight-like effects for localizing stimuli. However, no difference was found for upright and inverted faces in blindsight-like trials, which indicates that the face inversion effect diminishes at an unconscious level. Reported visibility revealed a bimodal distribution, faces were either completely visible or invisible. In conclusion, evidence for blindsight-like detection was found under b-CFS. However, no evidence was found for high-level face processing at zero subjective awareness.
3 Introduction
Faces are perhaps the most important visual stimuli that we perceive. Just from a single gaze, we can determine someone’s identity and emotional state. From birth to adulthood, humans preferably look at upright faces compared to other common objects
(Valenza, Simion, Cassia, & Umiltà, 1996). It seems that faces could therefore receive special processing in the brain. Indeed, evidence for this special processing is found in the fusiform face area that exclusively responds to face stimuli (Kanwisher, McDermott & Chun, 1997).
Another well-known example of possible special face processing refers to the fact that objects are harder to recognize when they are presented inverted compared to an upright presentation (Yin, 1969). Face recognition is disproportionally affected by this inversion effect, implying that inverted faces are not processed the same way as upright faces are. This face inversion effect has been used in many studies as a marker to indicate the possibility of specialized face recognition mechanisms. The holistic shape representation of faces is possibly responsible for this effect. When subjects had to identify a face based on parts (decomposition of the face), the inversion effect was eliminated (Farah, Maxwell & Tanaka, 1995). Also, faces were poorly recognized when isolated features of the face were presented compared to when the whole face was presented (Tanaka & Farah, 1993). This indicates that the processing of faces is associated with holistic pattern perception and differs from the processing of other common objects.
Interestingly, this inversion effect is not limited to recognition only, but also emerges during detection tasks (Purcell & Stewart, 1988). An upright face was more easily detected than a pattern of arbitrarily rearranged facial features or an inverted face. This means that the configuration of a face even influences detectability before subjects could recognize the face. Thus, the underlying mechanisms of face recognition are probably influenced by a specific face configuration that influences detection.
Previous research indicated that faces receive enhanced processing in recognition and detection tasks under subjective awareness. This implies that we consciously process faces different from non-facial stimuli. It remains however unknown if specialized face processing mechanisms are active without visual awareness and if the face inversion effect still emerge at an unconscious level.
One of the first paradigms introduced to study unconscious processing of visual stimuli is binocular rivalry. This paradigm refers to the spontaneous alternation in perception when two different stimuli are presented dichoptically in each eye (Jiang, Costello, He, 2007). It is believed that two stimuli are competing for visual awareness, in which the perceptual
4 switching indicate which stimulus is dominant at that time. A variation of binocular rivalry is continuous flash suppression (CFS). During CFS, a flashing high-contrast pattern is presented in one eye, while another stimulus of interest is presented in the other eye. CFS is especially interesting in comparison with binocular rivalry, because subliminal information can be presented for a longer time due to the suppression of the flashing pattern. Recent
neuroimaging studies have demonstrated that some high-level information is processed during interocular suppression, such as that emotional faces generate stronger responses in the
amygdala than neutral faces during interocular suppression (Williams, Morris, McGlone, Abbott, & Mattingley, 2004).
Recently, a new variation of CFS is often used for studying the process of visual information in the absence of visual awareness. This so called breaking continuous flash suppression (b-CFS) was first introduced in a study by Jiang, Costello and He (2007), which found that upright faces took less time to get access to awareness than inverted faces in b-CFS. This suggests that upright faces have an advantage in gaining consciousness compared to inverted faces. In addition to faces, evidence of inversion effects for human bodies during interocular suppression were found in b-CFS (Stein, Sterzer, & Peelen, 2012).
The time it takes for an interocular suppressed stimulus to ‘break through’ and become visible is commonly interpreted in b-CFS as a measurement of access to awareness (Gayet, Van der Stigchel & Paffen, 2014). The difference in break-through time between stimuli reflects to the potency to gain visual awareness in b-CFS. However, often an inference is made about unconscious processing of visual stimuli, in the sense that a shorter ‘break-through’ time for a stimulus reflects enhanced unconscious processing, which leads to faster access to awareness. In other words, by the response time to visible stimuli an inference is made about unconscious processing of invisible stimuli.
b-CFS studies that make this inference about unconscious processing of stimuli are difficult to interpret. It is not directly tested if faster ‘break through’ time indeed indicates enhanced unconscious processing. Instead, these results are based on assumptions. A possible way to study this more directly is done by Vieira, Wen, Oliver and Mitchell (2017). This study found that fearful faces, in contrast with other face expressions, showed enhanced localization differences, even when the faces were invisible. Subjects had to localize stimuli and indicate how confident they were for seeing the stimuli during b-CFS. They showed that when an objective as well as a subjective measurement is included, it is possible to directly examine if localization is enhanced at subjective visual unawareness. This enables the possibility to say something about the unconscious processing of emotional faces.
5 This ‘above chance localization’ resembles the effects of blindsight. The phenomenon blindsight refers to the ability of patients with cortical blindness to localize and detect visual stimuli that they claim to not have seen (Cowey, 2009). In trials where subjects did not see the stimulus, but performed above chance level in a localization/detection task are therefore referred to as blindsight trials.
Unfortunately, research such as Vieira et al. (2017) is sparse. Other studies claim to have tested unconscious processing, while only inferences were made. In this study, it is examined if the findings of Vieira et al. (2017) extend to localization differences for emotional neutral faces (upright vs inverted faces) in b-CFS for visible and invisible face stimuli. On every trial, the localization accuracy is measured as well as a visibility rating of the faces. By including upright and inverted faces as stimuli, it can be examined if the inversion effect is present during absence of subjective awareness and if face localization overall is above chance level at subjective unawareness.
We expect that upright faces are better localized than inverted faces under b-CFS. This would propose that the face inversion effect emerged under b-CFS. If no special face
processing occurs at an unconscious level, we expect that no differences will be found for localization accuracy for inverted and upright invisible stimuli. This would propose that the face inversion effect diminishes at zero subjective awareness. We still would expect that faces overall, will have a localization accuracy above chance, reflecting a ‘blindsight-like’ effect. However, if special face processing occurs at unconscious level, we expect that upright invisible faces are localized better than inverted invisible faces.
The subjective measurement that is included in our study, can be operationalized either in a discrete or in a continuous fashion. Additional to the study of Vieira et al. (2017), a continuous scale to rate the visibility level of the participants was included in this experiment. This was included because it provides an useful tool to reveal whether consciousness is a gradual or all-or-none phenomenon (Sergent & Dehaene, 2004). The nature of consciousness is a well discussed topic. Some theories support the idea that consciousness is the result from a wave of neural activity that tips over the threshold for brain activity to become
consciousness (Dehaene, 2014). Other theories, such as the signal detection theory and connectionist models, suggests that consciousness is gradual (Kanwisher, 2001). In an attentional blink task, participants were asked to rate the visibility of the second target (T2) (Sergent & Dehaene, 2004). They rated the visibility on a continuous scale, which allowed to examine whether attentional blink is a gradual of an all-or-none phenomenon. The results indicated that attentional blink has a bimodal distribution. T2 was either entirely visible or
6 invisible and thus not gradual. Adding a continuous scale as described in the research of Sergent and Dehaene (2004) has never been done in a b-CFS study. Therefore, it seems interesting to include this in this study. By looking at the distribution of the participants’ responses on the continuous scale, it can be concluded if visibility is a gradual process or not.
Materials and methods Ethic statements
A written informed consent was obtained from all participants. Participants
Fourteen participants above the age of 18 were recruited through the University of Amsterdam. All participants had normal or corrected-to-normal vision and were in good health. Participants received either 1.00 study credits or 8 euros as compensation. The participants were informed about the purpose of the experiment.
Apparatus and stimuli
Participants viewed the monitor (100Hz refresh rate, Dell Monitor) dichoptically through a custom-built mirror stereoscope. Their heads were stabilized with a chin-and-head rest at a viewing distance of 50 cm from the screen. The display showed two boxes with a noise border (8 pixels wide on both sides) and an inner medium gray area (284 pixels x 284 pixels). These boxes were fused together when the participant looked through the mirror stereoscope. The center of the box (fixation point) was 283 pixels away from the black background of the screen. The fixation point was presented 1s before the beginning of every trial, then it again disappeared for 400ms. The participants were instructed to maintain stable fixation during the experiment. The experiment took place in a dark room with dimmed lights. This setting was stable throughout the whole experiment.
The test stimuli for the adjustment task included scrambled faces, in which elements of faces were cut into boxes and rearranged. Inverted or upright faces were presented during the main task. The faces were proximally 72 pixels x 72 pixels and 42 pixels away from the fixation point at either left or right. Sixteen different faces were presented, each face was presented twice; upright/inverted and left/right for three different presentation times. By including upright and inverted faces, low-level image properties remained the same during all
7 the trials. Hereby, differences between trials can be assigned to high-visual processing (Gayet, Van der Stigchel and Paffen, 2014).
Procedure
Adjustment task. Before testing, the eye dominance was established for each participant. This was done with the distance-hole-in-card test. A participant had to extend their arms parallel in front of them, while holding a paper with a hole in it. The instructor held an object in front of the participant (about 3 meters distance) and instructed to look through the hole at the object. The participant then had to close one of their eyes. The opened eye that could still see the object, was considered as the dominant eye.
The adjustment procedure was used to test how accurate the participants scored on the task and to prevent learning effects. This task contained only scrambled faces, that were presented for one fixation time (0.4 s) in the dominant eye. After the stimulus was presented, the participants had to localize the scrambled face. They had to indicate this with < or > on the keyboard. The accuracy on this task indicated how well the participants performed in localizing scrambled faces. In order to prevent participants to be too good in this task, the target eye was switched from dominant eye to non-dominant eye when the accuracy was 70% or above. This means that during the main task, the target was presented into the
non-dominant eye instead of the non-dominant eye. All of the participants were informed that the task was difficult and that the scrambled faces were difficult to detect.
b-CFS continuous scale task. The main task consisted of 384 trials, where in each trial an upright or inverted face was presented. The presentation times for the faces were either 0.2s, 0.4s or 0.8s. Over the first 200ms of each trial, the face was faded in into the dominant eye (or non-dominant eye with adjustment task accuracy of 70% or above). At the beginning the contrast was zero, this contrast increased in 20 steps over the span of the presentation time. A flashing Mondrian like image was presented in the non-dominant eye (determined by the hole-in-cart test), with a flash at every 100ms (figure 1). In order to prevent facial after images, 3 masks were presented in both eyes after the stimulus appeared in each trial. One of the participants had an accuracy of 85% on the adjustment task, the target was presented in the non-dominant eye for the b-CFS continuous scale task.
8 After the stimulus presentation, the participants had to localize the targets either left or right of the box. This was done by pressing < or > on a keyboard. Following up, a continuous scale was presented. Subjects had to indicate how confident they were about spotting the targets (from invisible to visible). With a mouse, they had to slide on a continuous scale from not visible at all on the left side to completely visible on the right side. During the main task, three mandatory breaks were included (at every 96 trials), in which the participant could rest for at least 10 sec. All the participants were informed that their response was not time
pressured during the main task and the adjustment task. Figure 1
Face stimuli, upright presentation (left) and Mondrian like flashing pattern (right)
R esul ts A ll the data of the 14 participants were included in the analysis. The statistical analyzes were all
conducted in IBM SPSS Statistics, version 24 (IBM Corp).
Participants were overall more accurate in localizing upright faces rather than inverted faces for all three presentation times (figure 2). A repeated measures ANOVA was performed with two within-subject variables (upright vs inverted faces and short, medium and long presentation time). A main effect was found for face orientation, F(1,13)=11.84, p=.004 and presentation time, F(2,26)=11.57, p<.0.001. No interaction effect was found between face orientation and presentation time F(2,26)=1.07, p>0.05. These results indeed indicated that upright faces were localized better than inverted faces, which supports our expectations. Participants were also more accurate as the presentation time increased.
9 Figure 2
Mean localization accuracy (%) of upright and inverted faces for short, medium and long presentation time
The continuous scale as divided into proportions. Hereby, 0 indicated that the participants saw no faces and 1 indicated that they saw a face. The mean was calculated for upright vs inverted faces with the three presentation times (figure 3).
A repeated measures ANOVA was performed to test if an upright face orientation was subjectively scored higher than an inverted face orientation for the three presentation times. The results indicated a main effect of face orientation, F(1,13)=6.02, p=.029, as well as a main effect for presentation time, F(2,26)=4.81, p=0.017. Also, an interaction effect was found between face orientation and presentation time, F(1.086, 14.12)=8.97, p=.009. These results are according to Greenhouse-Geisser’s correction, since the assumption of sphericity was violated. Furthermore, three comparisons were made, with a paired sample t-test, in order to locate the mean differences. The p-value was adjusted to 0.05/3=0.017. Upright faces vs inverted faces for the short presentation time revealed no significant difference (t(13)=2.39, p=.033). However, a significant effect was found for upright vs inverted faces for medium presentation time (t(13)=2.88, p=.013) and long presentation time (t(13)=2.86, p=.013). These results indicated that for the shortest presentation time (0.2s) face orientation did not affect the rate of visibility in localizing the stimuli. Remarkably, the visibility rating increased as a straight line over all the presentation time, while the localization accuracy seemed to have
10 raised prominently between short and medium presentation time and less between the medium and long presentation time.
Figure 3
Mean visibility rate from 0-1 for upright and inverted faces; short, medium and long presentation time
Faces that were rated on the lowest 10% on the continuous scale, are considered as invisible faces. Notice that this is a quiet arbitrary decision and could influence the results. Figure 4 revealed that invisible faces (either upright or inverted) have a mean accuracy above chance level (M=0.58, SD=0.89). A one sample t-test was performed to test if faces were localized significantly above chance level under zero subjective awareness. The results indicated that invisible faces are localized above chance level (t(13)=3.56, p=.003). These results are in line with our expectations. A paired samples t-test was included to compare the mean accuracy of invisible upright and invisible inverted faces. This indicated that there is no significant effect of face orientation at subjective unawareness (t(13)=1.52, p>.05). No
evidence is here found for a face inversion effect at subjective unawareness, which is not in line with our expectations.
11 Figure 4
Mean accuracy of upright faces (invisible and visible) and inverted faces (invisible and visible).
Finally, it was tested if the proportion CS (continuous scale) presses showed a bimodal distribution or an uniform distribution. Figure 5 revealed that the visibility of the face stimuli is likely a to be a bimodal distribution; the most presses were made in the first and last bin of the CS. It seems that the most presses in the first bin were made in the shortest presentation time and decreases as the presentation time increases. This is in contrast with the last bin of the CS; a higher proportion of participants pressed in the last bin of the CS for the long presentation time and this amount decreases as the presentation time decreases (table 1)
12 Figure 5
The mean proportion of CS presses in each bin of the CS for short, medium and long presentation time
Table 1
Mean proportion (standard deviation between brackets) CS presses in the first and last bin for three presentation times.
Presentation time First bin Last bin
Short (0.2s) 0.40(0.28) 0.064(0.12)
Medium (0.4s) 0.36(0.27) 0.11(0.20)
Long (0.8s) 0.31(0.24) 0.16(0.26)
Discussion
In this study, it was examined whether the face inversion effect emerges under b-CFS and whether faces are localized above chance level at zero subjective awareness. Additional to that, it was examined whether the inversion effect was still present under zero subjective awareness. A continuous scale was included in this study since previous research from
13 Sergent and Dehaene (2004) showed that a continuous subjective scale is a useful for testing whether visibility is an all-or-none or gradual phenomenon.
The results indicated that upright faces are better localized than inverted faces for all presentation times. It can therefore be concluded that the face inversion effect is present under b-CFS. Also, participants rated the visibility of upright faces higher than inverted faces for medium and long presentation time. However, this was not found for the short presentation time. It seems that face orientation does not affect visibility level at a presentation time of 0.2s. This could indicate that awareness of visual stimuli takes time to emerge. Regarding underlying neural mechanisms, these results indicate that when face stimuli are presented for 0.2s, it is processed at such a level that the face inversion effect is present, but is not
reportable in terms of a visibility rating. Another remarkable result revealed that the slope of the mean localization accuracy showed a slight curve, while the mean visibility rate showed a linear line. In other words, subjective values and objective values do not correspond. While the face localization accuracy showed a bigger increase between the short and medium presentation, the visibility rating increased linearly. Also, an interaction effect was found for presentation time and visibility; upright faces were rated as more visible than inverted faces, this difference increased as the presentation time increased. In terms of localization
differences, upright and inverted faces did not have an interaction effect. This again indicates that participants responded in a linear fashion on the visibility rating, but this was not the case for localization accuracy. This discrepancy indeed could indicate that face processing is already emerging before subjects could report faces.
Further findings indicated that invisible faces (upright and inverted) were localized above chance level, reflecting ‘blindsight-like’ effects. However, no effect was found for an advantage of invisible upright faces over invisible inverted faces. This would mean that the face inversion diminishes at zero subjective awareness. Unconscious processing of face stimuli does not make a distinction in face orientation. This could indicate that faces are processed as common objects at an unconscious level. In contrast, Vieira et al. (2014) implied that localization differences between emotional neutral faces and fearful faces were present under subjective unawareness. A possible explanation for their finding is that ecological relevant stimuli breaks faster through interocular suppression (Gayet, Van der Stigchel & Paffen, 2014), such as emotional face expressions (Vieira et al., 2014) and eye contact (Stein, Senju, Peelen & Sterzer, 2011), than less ecological relevant stimuli. Upright and inverted faces lack emotional expressions and are therefore perhaps less ecological relevant and do not show enhanced localization at subjective unawareness. However, notice that in this study we
14 regarded ‘invisible’ as the lowest 10% of the CS. By changing this criteria to a wider range, non-significant outcomes could become significant.
Another finding suggests that visibility has a bimodal distribution. Faces were either completely visible or invisible. This is evidence for the all-or-none conscious theories, such as the global workspace theory. This theory suggests that consciousness is brain-wide
information sharing. Competing information tries to reach the global workspace, this information becomes conscious when it enters the global workspace (Dehaene, 2014).
These findings are however difficult to interpret, since one of the main disadvantages of this study regards the small number of subjects that participated. Power analysis (Matlab R2017b) showed an estimated power of 0.4102 with 14 participants and 0.5 Cohen’s effect size. This power is not a sufficiently high enough to correctly reject or assume a hypothesis. A minimum of 34 participants is needed to obtain a power of 0.8, which is considered as a high power. By including more participants, some results could become significant or
non-significant. A slightly higher localization accuracy was found for invisible upright vs invisible inverted faces, this indicates that this result could be significant for a larger number of
participants.
A second shortcoming of this study refers to the fact that the intermediate part of continuous scale is more vaguely defined than the outer parts. Participants had the clear instructions to press the most left side of the scale when they did not see a facial and the most right side when they did see a face. The intermediate part of the scale was pressed when participants had a hinge that a face could have been presented, but this ‘hinge’ is not a strict principle. Criteria differences for reporting a hinge vs seeing a face completely are dependent of the participant. This shortcoming was also noticed by Overgaard, Rote, Mouridsen, and Ramsøy, (2006) in evaluating the continuous scale used by Sergent, and Dehaene (2004). As only the extremes of the scale are labeled with descriptions, responses between these are ambiguous. Using more than four response categories can confuse the participants, therefore the 20 bins of the continuous scale is not likely to produce meaningful data. However, when the aim of a study is to measure consciousness, it is unavoidable to use subjective
measurements. Response bias will therefore always be an issue. Further research on this matter should include more participants (at least 34) and consider the methodological possibilities in subjective measurements.
Another issue regarding our experimental design refers to decrease of sustaining attention during the b-CFS task. Sustained attention is the capacity to maintain an accurate response over time across tasks. This decreases over time for monotonous and undemanding
15 tasks (Robertson, & O’Connell, 2010). Participants indicated that the task was boring and took effort to stay focused on. A decrease of focus results in well-known performance
decrement over time (Langner, & Eickhoff, 2013). Localization accuracy could therefore have suffered over time because of this effect.
Remarkably, Stein, Reeder, and Peelen (2016) showed that inversion effects in
conscious access under b-CFS are not limited to faces only, but also exists for other objects of expertise, such as cars. The face inversion could therefore reflect perceptual expertise instead of face-specific mechanisms. For further research, it would be interesting to test whether inversion effects of cars or other expertise categories still exist at zero subjective awareness. If these inversion effects are also abolished under subjective unawareness, it is likely that this is modulated by shared perceptual expertise mechanisms.
Altogether, blind-sight like effects are found for localizing invisible faces under b-CFS. Previous research indicated that faces are specially processed at subjective awareness from other common objects, but is it uncertain if this reaches an unconscious level.
16 References
Cowey, A. (2010). The blindsight saga. Experimental brain research, 200(1), 3-24. Dehaene, S. (2014). Consciousness and the brain: Deciphering how the brain codes
our thoughts (pp. 161-165). New York, New York: Penguin.
Farah, M. J., Tanaka, J. W., & Drain, H. M. (1995). What causes the face inversion effect? Journal of Experimental Psychology: Human perception and
performance, 21(3), 628.
Farah, M. J., Wilson, K. D., Drain, M., & Tanaka, J. N. (1998). What is" special" about face perception? Psychological review, 105(3), 482.
Gayet, S., Van der Stigchel, S., & Paffen, C. L. (2014). Breaking continuous flash
suppression: competing for consciousness on the pre-semantic battlefield. Frontiers in
Psychology, 5, 460.
Jiang, Y., Costello, P., & He, S. (2007). Processing of invisible stimuli: Advantage of upright faces and recognizable words in overcoming interocular suppression. Psychological
science, 18(4), 349-355.
Kanwisher, N. (2001). Neural events and perceptual awareness. Cognition, 79(1-2), 89-113. Kanwisher, N., McDermott, J., & Chun, M. M. (1997). The fusiform face area: a module in
human extrastriate cortex specialized for face perception. Journal of
neuroscience, 17(11), 4302-4311.
Langner, R., & Eickhoff, S. B. (2013). Sustaining attention to simple tasks: A meta-analytic review of the neural mechanisms of vigilant attention. Psychological bulletin, 139(4), 870.
Purcell, D. G., & Stewart, A. L. (1988). The face-detection effect: Configuration enhances detection. Perception & Psychophysics, 43(4), 355-366.
Overgaard, M., Rote, J., Mouridsen, K., & Ramsøy, T. Z. (2006). Is conscious perception gradual or dichotomous? A comparison of report methodologies during a visual task. Consciousness and cognition, 15(4), 700-708.
Robertson IH, O’Connell RG. Vigilant attention. In: Nobre AC, Coull JT, editors. Attention and time. Oxford, UK: Oxford University Press; 2010. pp. 79–88.
Sergent, C., & Dehaene, S. (2004). Is consciousness a gradual phenomenon? Evidence for an all-or-none bifurcation during the attentional blink. Psychological science, 15(11), 720-728.
17 objects of expertise. Journal of Experimental Psychology: Human Perception and Performance, 42(6), 788.
Stein, T., Senju, A., Peelen, M. V., & Sterzer, P. (2011). Eye contact facilitates awareness of faces during interocular suppression. Cognition, 119(2), 307-311.
Stein, T., Sterzer, P., & Peelen, M. V. (2012). Privileged detection of conspecifics: Evidence from inversion effects during continuous flash suppression. Cognition, 125(1), 64-79. Tanaka, J. W., & Farah, M. J. (1993). Parts and wholes in face recognition. The Quarterly
journal of experimental psychology, 46(2), 225-245.
Valenza, E., Simion, F., Cassia, V. M., & Umiltà, C. (1996). Face preference at birth. Journal
of experimental psychology: Human Perception and Performance, 22(4), 892.
Vieira, J. B., Wen, S., Oliver, L. D., & Mitchell, D. G. (2017). Enhanced conscious
processing and blindsight-like detection of fear-conditioned stimuli under continuous flash suppression. Experimental brain research, 235(11), 3333-3344.
Williams, M. A., Morris, A. P., McGlone, F., Abbott, D. F., & Mattingley, J. B. (2004). Amygdala responses to fearful and happy facial expressions under conditions of binocular suppression. Journal of Neuroscience, 24(12), 2898-2904.
Yin, R. K. (1969). Looking at upside-down faces. Journal of experimental psychology, 81(1), 141.
