We have tested whether natural dynamic backgrounds affect perceived facial dominance. Facial evaluation is based on just two fundamental dimensions of facial appearance: valence and dominance. Perceived facial valence has been shown to be biased towards background valence. However, it is currently unknown, if the perception of facial dominance is also context dependent. In this study, participants rated the perceived dominance of neutral faces superimposed on everyday dynamic backgrounds that were classified as either low (weak) or high (strong) in dominance. Neutral faces were perceived as significantly more dominant when seen on a strong dynamic background than on either a weak or neutral background. Thus, background dominance enhances perceived facial dominance. Since dynamic textures are ubiquitous this finding is relevant for the design and experience of both our daily environment and multimedia content.

People are typically unaware that the context in which they see a face influences their affective judgement, even when this context has absolutely no relevance for their assessment^{[4][5][6][7][8]}. Associations between context and faces are routinely established and modulate face perception already at the early stages of facial feature processing, such that the affective quality of the context transfers to the perceived affective state of a face^[4][8]. As a result, identical facial configurations may convey strikingly different emotions and dimensional values depending on the context in which they are perceived^[5]. Previous studies shown that perceived facial valence is biased towards background valence; faces perceived in negative (or positive) contexts are judged to be more negative (or positive) than in a neutral context^[5][6]. However, it is currently unknown if the perception of facial dominance is also context dependent. Environmental psychology has shown that the dominance is one of the principal affective qualities of backgrounds and environments. The dominance (sometimes also called strength or potency) of a background is defined as the degree to which it affects the observer^[9][10]. An observer may feel overwhelmed by, and unable to control, a strong background, while (s)he may feel in control of, and able to influence, a weak background media. We hypothesised that background dominance may also biased perceived facial dominance. To test this hypothesis, we performed an experiment in which observers judged the dominance of neutral faces presented on natural dynamic textures (spatially repetitive, time-varying visual patterns that repeat, or seems to repeat, themselves over time;^[11] these were classified as neutral, low (weak) or high (strong) in dominance (Fig. 1A).

The objective of this study was to investigate whether perceived facial dominance is biased by the dominance of everyday dynamic backgrounds.

The results of the experiment were first accumulated by calculating the median dominance rating per background across all 12 faces for each of the 30 participants (Fig. 1B and C). Next, these median dominance ratings were accumulated across the 6 strong backgrounds and across the 6 weak backgrounds individually (again by using the median). These ratings were then compared with each other and with neutrality (i.e. zero). A Friedman test revealed a significant effect of background on dominance rating ($chi^2$ = 13.972, p = 0.001). Next, a post-hoc analysis was performed using multiple Wilcoxon singed rank tests, which were conducted using Bonferroni-adjusted alpha levels of 0.0167 per test (0.05/3). The results show a significant difference between perceived dominance ratings of faces on strong and weak backgrounds respectively (Z = 3.033, p = 0.002). Also, perceived dominance ratings for faces on strong backgrounds are significantly different from neutral ones (Z = 3.305, p = 0.001). However, perceived dominance ratings for faces on weak backgrounds are not significantly different from neutral ones (Z = 2.230, p = 0.026). Neutral faces are perceived as significantly more dominant when seen against a strong background than on either a weak or a neutral (dark) background. A Mann-Whitney U test was performed to test for gender difference. For neutral faces shown on strong backgrounds this test revealed no significant difference between dominance ratings by males (Md = 0.32, N = 18) and females (Md = 0.23, N = 12), U = 86.5, Z = -0.911, p = 0.368, r = -0.16. For neutral faces shown on weak backgrounds there was also no significant difference between males (Md = 0.08, N = 18) and females (Md = 0.12, N = 12) , U = 92.0, Z = -0.678, p = 0.518, r = -0.12.

Our current finding that even everyday dynamic textures such as streaming water, swirling leaves, moving clouds, waving flags or traffic streams can influence perceived facial dominance agrees with the growing body of evidence that background context modulates perceived facial emotions^[5][6][7]. The current results also agree with recent brain studies showing that contextual information influences activities in the extended neural network of face processing and thus alters the perception and evaluation of facial expressions^{[12][13][8][14]}. In particular, it has recently been observed that the amygdala integrates facial expression with salient motion information^[15]. This indicates that the amygdala is not only is responsive to facial expressions^[16] but also incorporates the overall perceptual context of a stimulus. Hence, it appears that facial evaluation is context dependent and not automatic, hard-wired, effortless and universal as previously proposed^[17].

In addition to their high prevalence in our daily environment, dynamic textures are also increasingly applied in animation^[18] and video synthesis^[11], and are deployed on large-scale digital billboards and electronic wallpapers^[19]. Because of this ubiquity of dynamic textures, our current findings may be relevant for the design and experience of both our daily environment and multimedia content.

Everyday dynamic background textures such as streaming water, swirling leaves, moving clouds, waving flags or traffic streams can bias perceived facial dominance.

A limitation of this study is the small number of stimuli (faces and backgrounds) that were tested. Future studies should use a larger number of dynamic background textures with widely varying content and motion patterns, covering the entire dominance range, to allow a closer investigation (correlation) of the relation between background dominance and perceived facial dominance. In addition, artificial affective motion textures with well specified path curvature, speed and texture layout^[18] may serve to systematically investigate the relation between different spatio-temporal texture parameters and perceived facial dominance.

We expect that visual background dominance may also bias the affective appraisal of non-face objects with no evident semantic affective connotation. Moreover, it is also likely that this effect may carry over to other sensory modalities^[20]. Hence, a dominant visual background may bias the perception of certain smells, tastes or sounds. We plan to investigate these issues in a follow-up study.

30 (12 female) observers (ages 21–64 years, mean age 40 years) rated the dominance of 12 neutral faces overlaid on 12 different dynamic backgrounds, using a 5 point rating scale ranging from very submissive to very dominant. 6 of these backgrounds have previously been classified as high in dominance (strong backgrounds: mean dominance rating 0.85±0.07 on a scale from 0 to 1, N = 35^[3]) while 6 others have been classified as low in dominance (weak backgrounds: mean dominance rating 0.22±0.06, N = 35^[3]).

The 12 neutral faces were selected from the Dominance data set of the validated Princeton faces database^[1]. Their neutrality had been verified in a previous study for a homogeneous dark background^[22]. Neutral faces were used as targets since their evaluation is affected by emotional scene content to a greater extent than the evaluation of faces with exaggerated facial expressions^[5][6][8], probably because of their ambiguous nature^[23][24]. In addition, by using neutral faces, issues of stimulus-background congruency can be avoided^[5].

The 12 dynamic backgrounds were different natural textures from the Dyntex database (AVI movies with a resolution of 600×480 pixels, a duration of 10 s, and a frame rate of 25 fps^[2]), representing everyday background scenes like moving water, fluttering vegetation and a waving flag. In an earlier study^[3] 6 of these textures (with identifiers 54ab110, 64adl10, 648dc10, 649ha10, 6484d10, and 6485110 in the Dyntex database) were classified as high in dominance (strong), and 6 others (with identifiers 54ac110, 571b110, 645ab10, 6486b10, 6482210, 6485310) were classified as low in dominance (weak; Fig. 1B).

In contrast to the previous studies on the effects of affective backgrounds on facial evaluation^[5][6] the backgrounds used in this study are dynamic and have no evident semantic affective connotation. Each face was overlaid on each dynamic background, resulting in a total of 144 different stimuli (12 faces×12 backgrounds).

Dell Precision 490 PCs were used to present the stimuli to the observers in random order and to register their response. The computers were equipped with Dell 19" monitors, with a screen resolution of 1280×1024 pixels, and a screen refresh rate of 60 Hz. MediaLab v2012 (www.empirisoft.com) was used to present the stimuli and collect the answers. The stimuli were presented for maximally 10 s on a light grey background, flanked by a rating scale. If a participant responded within 10 s after the onset of a stimulus presentation, the current face would disappear and the next face would be shown. If a participant needed more than 10 s to respond, the stimulus disappeared from the screen but the rating scale remained visible until the participant had responded. Participants were instructed to base their answers solely on their first and overall impression of each face and to ignore the background. Observers used standard mouse pointers to indicate their response. Statistical analyses were performed with IBM SPSS 20.0 for Windows. As the experiments used an ordinal scale of measurement, without assuming it is also an interval scale, non-parametric tests were used.

The authors declare no conflicts of interest.

The participants read and signed an informed consent prior to the experiment. The experimental protocol was reviewed and approved by the TNO Ethics Committee and was in accordance with the Helsinki Declaration of 1975, as revised in 2013^[21].

No fraudulence is committed in performing these experiments or during processing of the data. We understand that in the case of fraudulence, the study can be retracted by ScienceMatters.