Chinese Aesthetic Mask: Three Forehead and Five Eyes—Holistic Processing and Facial Attractiveness

Abstract

Human face processing has been attributed to holistic processing. Here, we ask whether humans are sensitive to configural information when perceiving facial attractiveness. By referring to a traditional Chinese aesthetic theory—Three Forehead and Five Eyes—we generated a series of faces that differed in spacing between facial features. We adopted a two-alternative forced-choice task in Experiment 1 and a rating task in Experiment 2 to assess attractiveness. Both tasks showed a consistent result: The faces which fit the Chinese aesthetic theory were chosen or rated as most attractive. This effect of configural information on facial attractiveness was larger for faces with highly attractive features than for faces with low attractive features. These findings provide experimental evidence for the traditional Chinese aesthetic theory. This issue can be further explored from the perspective of culture in the future.

Keywords

facial attractiveness holistic processing configural processing inversion effect composite face effect Three Forehead and Five Eyes Marquardt golden ratio

When seeing a face, we process all face regions obligatorily (for reviews, see Hayward et al., 2013; Maurer et al., 2002; Richler & Gauthier, 2014; Rossion, 2013). Previous work has demonstrated such holistic processing when we perceive facial identity (Hancock et al., 2000), emotion (Calder et al., 2000), gender (Baudouin & Humphreys, 2006; Zhao & Hayward, 2010), and race categorization (Michel et al., 2007).

People may also rely on holistic processing when perceiving facial attractiveness (e.g., Abbas & Duchaine, 2008; Leder et al., 2017). For example, Abbas and Duchaine (2008) used the most popular paradigm of holistic face processing——the composite face paradigm (see Richler & Gauthier, 2014 for a review)——to investigate facial attractiveness. They found that the same top halves of upright faces were judged more attractive when they aligned with an attractive bottom half than when they aligned with an unattractive bottom half but did not observe this effect when faces were inverted or misaligned. Inversion, which has been well documented to interfere with holistic face processing (e.g., Yin, 1969; for reviews, see Burke & Sulikowski, 2013; McKone et al., 2013), can also affect facial attractiveness. Faces were rated more attractive when they rotated 90° or 180° than when they were upright, and the less attractive the faces were in general, the larger were the effects (Leder et al., 2017).

In the present study, we manipulated the spacing between facial features (i.e., configural information) to investigate the holistic processing in facial attractiveness perception. In other words, we ask whether humans are sensitive to configural information when perceiving facial attractiveness and whether there is an optimal spacing for attractiveness. Pallett et al. (2010) manipulated the distance between eyes and mouth. They found the optimal length ratio (0.36, dividing the eye-to-mouth distance by face length from hairline to chin) for the maximally attractive face. They then fixed this length ratio, manipulated the distance between the eyes, and found an optimal width ratio (0.46, dividing the pupils’ distance by face width between the ears’ inner edges). However, as they have discussed, since these ideal length and width ratios were obtained from only Caucasian female faces, there may be different optimal ratios for other races, males, or other ages.

In the present study, we are interested in how configural information affects Chinese faces’ attractiveness. Traditional Chinese aesthetics also attaches great importance of configural information to facial attractiveness. The concept of Three Forehead and Five Eyes (for short, 3F5E; see Figure 1 for illustration) was first coined in “The Secret of Portrait Painting” by portraitist Yi Wang in the late Yuan dynasty (cf. Yu, 1957). The 3F5E divided a face into three vertical parts with equal vertical distance and five horizontal parts with the same one-eye distance. Faces that meet this criterion are considered ideal.

Figure 1.

Standard spatial relations between features in the traditional Chinese aesthetics theory Three Forehead and Five Eyes.

The present study aims to investigate whether Chinese participants prefer faces with 3F5E. We adopted the best fixed-point in Zhang et al. (2017) to establish the faces met the standard 3F5E criteria. Zhang et al. (2017) analyzed 48 attractive Chinese faces to explore which fixed-point modes fit them best. According to their conclusion, Three Forehead (for short, 3Forehead) divides face height into three equal parts: from forehead hairline to eyebrow bone, from eyebrow bone to nose base, and from nose base to the bottom of chin; each takes 1/3 length of the face height. Besides, from nose base to edges of underlip and chin take 1/6 length of the face height. Five Eyes (for short, 5Eye) divides the face width into five equal parts: The length from the left ear’s edge to the right ear’s edge is equal to 5 times the length of one eye. Besides, the distance between two eyes is equal to the length of one eye, and the distance between the outside of two eyes and the edge of the ear is one eye, accounting for 1/5 of each (see Figure 1). Unlike Pallett et al. (2010) only manipulating one dimension distance at a time, we manipulated the vertical distance and the horizontal distance at the same time to explore whether the two dimensions are independent or not. We predicted that the faces which meet the criteria of 3F5E would be considered as most attractive.

We may use different strategies to assess the attractiveness of attractive and unattractive faces (e.g., Chang & Chou, 2009; Leder et al., 2017; Stróżak & Zielińska, 2019; Ying et al., 2020). For example, our visual system tends to extract high spatial frequency information when processing an unattractive face, while low spatial frequency information for an attractive face (Ying et al., 2020). The enhancement effect of inversion on facial attractiveness was more pronounced for low attractive faces than for highly attractive faces (Leder et al., 2017). High spatial frequency information was regarded as essential for encoding featural information, while low spatial frequency information captured the coarse configural information needed for holistic processing (Collishaw & Hole, 2000; Goffaux et al., 2005; Goffaux & Rossion, 2006). Inversion impairs the holistic processing of faces (e.g., Yin, 1969; see McKone et al., 2013 for a review). Taken together, we believe that attractive faces may be more sensitive to holistic processing, while unattractive faces are more sensitive to featural processing. Therefore, we manipulated the attractiveness of facial features to explore whether it modulates the 3F5E criterion. We predicted that faces with highly attractive features would be more sensitive to holistic processing.

In sum, our study manipulated the configural information and facial features’ attractiveness. We hypothesized that the faces which meet the criteria of 3F5E would be chosen (in Experiment 1) or be rated (in Experiment 2) as most attractive, and the effect of configural information was more significant for faces with highly attractive features.

Experiment 1

Method

Participants

Eighty Chinese college students (41 males, 39 females; age from 18 to 25, mean = 20.725) in Guangzhou participated in the study for payment. All participants had normal or corrected-to-normal vision. Half participants (21 males, 19 females, ages from 18 to 25, mean = 21.05) took part in Experiment 1a, and the other half participants (20 males, 20 females, ages from 19 to 25, mean = 20.40) participated in Experiment 1b.

Material

Eighty Chinese female faces from the internet, which had been rated in attractiveness and used in Liu (2020), were adopted as the original face materials in the present study. Previous studies have found that facial contour (e.g., Liu & Chen, 2018) and facial symmetry (e.g., Little et al., 2008; see Rhodes, 2006, for a review) affect facial attractiveness. To rule out their confounding, we generated a symmetric contour and applied it to all face materials. We selected one medium attractive face from the eighty Chinese female faces and mirrored its left side to obtain a left-right symmetric external face contour. As for the inner features, we selected some faces with high attractiveness ratings and some faces with low attractiveness ratings and used Photoshop to crop their eyes, nose, and mouth as highly/low attractive facial features.

We combined these inner features into the external contour mentioned above based on the criteria of 3F5E. We obtained eight standard Chinese female faces (four with highly attractive features and the other four with low attractive features). Based on the eight standard faces, we moved the inner features horizontally or vertically to create five levels of 5Eye and seven levels of 3Forehead (see Figure 2). A total of 35 feature spatial relations for each standard face made 280 grayscale pictures of Chinese female face stimuli. Each single face image extended a viewing angle of 5.55° × 7.00°. Besides, we cropped the left/right eyes (1.72° × 1.01°), nose (1.46° × 2.43°), and mouth (2.19° × 1.24°), retaining the size relationships among the whole face. In each trial, two faces displayed side by side, occupying 14.95° × 7.00° in total. The distance between the center of the two faces is 8.43°.

Figure 2.

Five levels of 5Eye (top), seven levels of 3Forehead (middle), and an illustration of the manipulation of vertical distance (bottom). E-standard = standard eyes position according to 5Eye; E-closer = move both eyes in standard position 2 pixels inward; E-closest = move both eyes in standard position 4 pixels inward; E-farther = move both eyes in standard position 2 pixels outward; E-farthest = move both eyes in standard position 4 pixels outward. S, M, and L mean to shorten, keep middle, and enlarge the specific 1/3 face height by 4 pixels, respectively. For example, F-LMS means the face has a larger “top forehead,” middle “middle forehead,” and a shorter “bottom forehead.” F-MMM means the standard vertical spatial layout of features according to 3Forehead. F-LSM means the face has a larger “top forehead,” shorter “middle forehead,” and a middle “bottom forehead.” Specifically, from nose base to underlip edges always takes 1/2 of the length from nose base to chin’s bottom.

Procedure

The experiment runs via E-prime software. Stimuli were presented on a screen with a resolution of 1,920 × 1,080 pixels on a black background. In each trial, after a 500 ms fixation at the center of the screen, two faces were presented symmetrically on the left and right sides of the screen until response. Each face was presented randomly on the left or right. Participants were asked to judge which one was more attractive. In Experiment 1a, in each trial, the two faces were the same in the 3Forehead and feature attractiveness level but different in the 5Eye levels, and the 3Forehead and feature attractiveness level varied between trials. There was a total of 560 trials in Experiment 1a, with a break between every 140 trials. Similarly, in Experiment 1b, the two faces always differed at 3Forehead but with the same 5Eye and same feature attractiveness in a trial. The latter two were manipulated between trials. There were 840 trials in experiment 1b, with a break between every 210 trials. We separate Experiment 1 into 1a and 1b in this way because there would be extensive trials if all of the 280 faces were paired as stimuli.

Besides, to check our manipulation of highly/low attractive facial features, we asked participants in Experiment 1a to rate the isolated facial features’ attractiveness after completing the forced-choice task. We adopted a 9-point Likert scale (1 = very unattractive, 9 = very attractive). In each trial, after a 500 ms fixation, a picture of an isolated facial feature was displayed at the screen center until participants’ response. Thirty-two facial feature pictures were presented randomly, one at a time.

Results and Discussion

A paired t test showed no significant difference between the left eye (M = 5.32) and right eye (M = 5.48), t(39)=1.55, p = . 130, Cohen’s d = 0.24. Therefore, the attractiveness ratings for the left eye and those for the right eye were averaged as the mean attractiveness ratings of the eyes. Then, three paired t tests were performed for the three isolated features (i.e., eyes, nose, and mouth), respectively. The results showed that highly attractive features received significantly higher ratings than low attractive ones. For eyes, M_high=6.59, M_low=4.22, t(39)=11.7, p<.001, Cohen’s d = 1.84. For nose, M_high=5.33, M_low=3.77, t(39)=7.31, p<.001, Cohen’s d = 1.15. For mouth, M_high=6.28, M_low=4.29, t(39)=8.42, p<.001, Cohen’s d = 1.33. These results validated the manipulation of feature attractiveness.

For the forced-choice task, the proportions of trials each face in each condition were chosen as more attractive were regarded as the dependent variable. We reported Greenhouse–Geisser corrected p values to counteract observed violations of sphericity.

In Experiment 1a, the data were submitted to a 5 (5Eye) $×$ 7 (3Forehead) $×$ 2 (Feature Attractiveness) repeated measure analysis of variance (ANOVA). Because the participants chose one from two faces identical in 3Forehead and feature attractiveness, neither the main effects of 3Forehead and feature attractiveness nor their interactions can be analyzed. Therefore, we focused on the main effect of 5Eye and its interaction with 3Forehead and feature attractiveness. The results showed a significant main effect of 5Eye, F (1.470, 569.442) = 11.592, p < .001, $η_{p}^{2}$ =.229. The interaction between 5Eye and feature attractiveness was also significant, F (1.668, 569.442) = 6.196, p = .006, $η_{p}^{2}$ = .133 (see Figure 3a). The results of simple effect analysis were shown in Table 1. For faces with highly attractive features, participants chose E-standard, E-closer, and E-farther the most times. For faces with low attractive features, E-standard and E-closer were selected the most. E-farthest and E-closest were considered the least attractive for both faces with high or low attractive features. These results suggest that merely changing the distance between the eyes can significantly alter facial attractiveness. Faces that met (i.e., E-standard) or were close (i.e., E-farther or E-closer) to the 5Eye criteria were the most attractive. Faces that deviated too much from this criterion were unattractive. By and large, these results were consistent with our hypotheses that the faces that meet the criteria of 5Eye would be regarded most attractive, which was modulated by the attractiveness of facial features. However, the interaction between 5Eye and 3Forehead did not reach significant, p = .395, $η_{p}^{2}$ = .026, indicating that the effect of 5Eye and 3Forehead were independent.

Figure 3.

Participants’ preference proportion as a function of feature attractiveness and 5Eye levels (a) in Experiment 1a and that of feature attractiveness and 3Forehead levels (b) in Experiment 1b.

Table 1.

The results of the simple effect analysis of the 5Eye × Feature Attractiveness interaction in Experiment 1a.

Levels		E-closest	E-closer	E-standard	E-farther	E-farthest
	High	0.405	0.573	0.605	0.549	0.368
	Low	0.473	0.585	0.584	0.506	0.353
E-closest	0.405		<0.001 (0.83)	0.003 (0.63)	0.251 (0.37)	1.000 (0.08)
E-closest	0.473		0.005 (0.60)	0.248 (0.37)	1.000 (0.09)	0.770 (0.28)
E-closer	0.573			1.000 (0.21)	1.000 (0.09)	0.005 (0.60)
E-closer	0.585			1.000 (0.01)	0.270 (0.36)	<0.001 (0.73)
E-standard	0.605				0.093 (0.43)	<0.001 (0.97)
E-standard	0.584				<0.001 (0.75)	<0.001 (1.08)
E-farther	0.549					<0.001 (1.12)
E-farther	0.506					<0.001 (1.00)
E-farthest	0.368
E-farthest	0.353

Note. The results are with Bonferroni correction. The values in parentheses are Cohen’s d. Significant results are in boldface. E-closest = move both eyes in standard position 4 pixels inward; E-closer = move both eyes in standard position 2 pixels inward; E-standard = standard eyes position according to 5Eye; E-farther = move both eyes in standard position 2 pixels outward; E-farthest = move both eyes in standard position 4 pixels outward.

Figure 4.

Participants’ attractiveness ratings as a function of 5Eye, 3Forehead, and feature attractiveness in Experiment 2.Note. Please refer to the online version of the article to view the figure in colour.

Similarly, in Experiment 1b, the data were submitted to a 5 (5Eye) $×$ 7 (3Forehead) $×$ 2 (Feature Attractiveness) repeated measure ANOVA. We only focus on the main effect of 3Forehead and its interaction with 5Eye and feature attractiveness. The results showed a significant main effect of 3Forehead, F (1.865, 554.734) =7.302, p = .002, $η_{p}^{2}$ = .158, and its interaction with feature attractiveness, F (2.580, 554.734) = 11.397, p < .001, $η_{p}^{2}$ = .226 (see Figure 3b). The simple effect analysis (Table 2) showed that the effect of 3Forehead for faces with highly attractive features was larger than that for faces with low attractive features. For faces with highly attractive features, F-MMM, F-LMS, and F-MLS were chosen the most while F-SML and F-SLM were selected the least. However, for faces with low attractive features, participants only chose F-MMM significantly more than F-SLM, and there was no difference between any two other conditions. These results were partly consistent with our hypothesis that the faces that meet the criteria of 3Forehead would be regarded most attractive. The effect of 3Forehead was larger for faces with highly attractive features. Interestingly, we also found that for faces with highly attractive features, faces with short chin (i.e., F-LMS and F-MLS) were also chosen as highly attractive. The interaction between 5Eye and 3Forehead did not reach significant, p = .167, $η_{p}^{2}$ = .032, consistent to Experiment 1a, again suggested that the effect of 5Eye and 3Forehead were independent.

Table 2.

The results of the simple effect analysis of the 3Forehead × Feature Attractiveness interaction in Experiment 1b.

		F-LMS	F-LSM	F-MLS	F-MMM	F-MSL	F-SLM	F-SML
Levels	High	0.602	0.532	0.537	0.544	0.477	0.413	0.394
Levels	Low	0.518	0.519	0.505	0.538	0.508	0.439	0.474
F-LMS	0.602		0.005 (0.64)	0.553 (0.37)	0.411 (0.39)	0.015 (0.58)	0.008 (0.62)	0.003 (0.67)
F-LMS	0.518		1.000 (0.00)	1.000 (0.09)	1.000 (0.13)	1.000 (0.06)	1.000 (0.31)	1.000 (0.17)
F-LSM	0.532			1.000 (0.03)	1.000 (0.10)	0.625 (0.36)	0.237 (0.42)	0.042 (0.52)
F-LSM	0.519			1.000 (0.08)	1.000 (0.14)	1.000 (0.08)	1.000 (0.31)	1.000 (0.19)
F-MLS	0.537				1.000 (0.06)	1.000 (0.32)	0.001 (0.75)	0.001 (0.74)
F-MLS	0.505				1.000 (0.32)	1.000 (0.02)	1.000 (0.42)	0.231 (0.18)
F-MMM	0.544					0.008 (0.62)	0.005 (0.64)	<0.001 (0.80)
F-MMM	0.538					1.000 (0.28)	0.006 (0.63)	0.176 (0.44)
F-MSL	0.477						1.000 (0.30)	0.035 (0.53)
F-MSL	0.508						0.670 (0.35)	1.000 (0.23)
F-SLM	0.413							1.000 (0.18)
F-SLM	0.439							1.000 (0.31)
F-SML	0.394
F-SML	0.474

Note. The results are with Bonferroni correction. The values in parentheses are Cohen’s d. Significant results are in boldface.

Experiment 2

Method

Participants

Another 30 Chinese college students (15 males, 15 females, ages from 18 to 25, mean = 20.31) in Guangzhou participated in the study for payment. All the participants had normal or corrected-to-normal vision.

Material

The materials were the same as those in Experiment 1.

Procedure

Participants were asked to rate the attractiveness of isolated features in the first block and that of intact faces in the second block on a 9-point Likert scale (1 = very unattractive, 9 = very attractive) in case the assessing of intact faces affects the assessment of features. There were 32 trials in the first block and 280 trials in the second block. There was a break every 70 trials in the second block. In each trial, a 1,000 ms fixation appeared first, and then the stimulus was presented at the screen center until response. All the stimuli were presented randomly, one at a time.

Results and Discussion

The mean attractiveness ratings in each condition were shown in Figure 4. First, a paired t-test showed no significant difference between the left eye and right eye (p =.143, Cohen's d=0.27). Therefore, the attractiveness ratings for the left eye and those for the right eye were averaged as the mean attractiveness ratings of the eyes. Then three paired t tests were performed for the three isolated features (i.e., eyes, nose, and mouth), respectively. The results showed that highly attractive features received significantly higher ratings than low attractive ones, for eyes, t(29) = 8.40, p<.001, Cohen’s d = 1.53; for nose, t(29)=5.66, p<.001, Cohen’s d = 1.03; for mouth, t(29) = 9.34, p<.001, Cohen’s d = 1.71. These results validated the manipulation of feature attractiveness.

Next, we conducted a 5 (5Eye) $×$ 7 (3Forehead) $×$ 2 (Feature Attractiveness) repeated measure ANOVA. Not surprisingly, the main effect of feature attractiveness was significant, F (1, 696) = 234.962, p<.001, $η_{p}^{2}$ = .890, further validating the manipulation of feature attractiveness. There was a significant main effect of 5Eye, F (2.082, 350.938) = 35.501, p<.001, $η_{p}^{2}$ = .550, and 3Forehead, F (6, 696) = 9.104, p<.001, $η_{p}^{2}$ = .239.

Consistent with Experiment 1a, there was a significant interaction between feature attractiveness and 5Eye, F (2.604, 350.938) = 16.996, p<.001, $η_{p}^{2}$ = .370. The results of the simple effect analysis (Table 3) showed that the effect of 5Eye for faces with highly attractive features was larger than that for faces with low attractive features. For faces with highly attractive features, E-standard and E-farther faces received the highest attractiveness rating. Meanwhile, E-farthest and E-closest were considered the least attractive. For faces with low attractive features, E-closest was considered the least attractive, while the other four levels did not differ significantly. Consistent with our hypothesis, for faces with highly attractive features, faces that met the 5Eye criteria were highly attractive, while faces with high deviations from the criteria were least attractive. Nevertheless, for faces with low attractive features, the effect of eye distance is not as large. However, we still observed that when the face deviated a lot from the 5Eye (i.e., E-closest), people considered it to be significantly less attractive. Interestingly, a little bit different from the E-closer in Experiment 1a, the faces with slightly farther eye distance (i.e., E-farther) than the standard 5Eye face were also considered very attractive.

Table 3.

The results of the simple effect analysis of the 5Eye × Feature Attractiveness interaction in Experiment 2.

		E-closest	E-closer	E-standard	E-farther	E-farthest
Levels	High	4.706	5.406	5.723	5.605	5.051
Levels	Low	2.250	2.431	2.557	2.520	2.411
E-closest	4.706		<0.001 (1.62)	<0.001 (1.57)	<0.001 (1.19)	0.145 (0.47)
E-closest	2.250		0.004 (0.74)	0.001 (0.82)	0.004 (0.72)	0.165 (0.46)
E-closer	5.406			<0.001 (0.90)	0.225 (0.44)	0.040 (0.57)
E-closer	2.431			0.219 (0.44)	1.000 (0.30)	1.000 (0.06)
E-standard	5.723				0.152 (0.47)	<0.001 (1.32)
E-standard	2.557				1.000 (0.14)	0.250 (0.43)
E-farther	5.605					<0.001 (1.03)
E-farther	2.520					0.669 (0.35)
E-farthest	5.051
E-farthest	2.411

Expectedly, the interaction between 3Forehead and feature attractiveness was also significant, F (4.674, 350.938) = 7.332, p<.001, $η_{p}^{2}$ = .202. The simple effect analysis (Table 4) showed that the effect of 3Forehead for faces with highly attractive features was larger than that for faces with low attractive features. For faces with highly attractive features, F-MMM, F-LMS, and F-MLS were rated as most attractive. For faces with low attractive features, except that F-MMM faces were significantly more attractive than F-SLM faces, there was no significant difference between any other faces. These results were highly consistent with Experiment 1b and partly consistent with our hypotheses. Besides faces that meet the criteria of 3Forehead, faces with short chin were also rated as very attractive in highly attractive features condition.

Table 4.

The results of the simple effect analysis of the 3Forehead × Feature Attractiveness interaction in Experiment 2.

		F-LMS	F-LSM	F-MLS	F-MMM	F-MSL	F-SLM	F-SML
Levels	High	5.535	5.217	5.373	5.512	5.222	5.073	5.155
Levels	Low	2.405	2.363	2.463	2.492	2.395	2.480	2.438
F-LMS	5.535		0.038 (0.62)	0.533 (0.43)	1.000 (0.07)	0.045 (0.61)	<0.001 (1.24)	0.001 (0.84)
F-LMS	2.405		1.000 (0.19)	1.000 (0.17)	1.000 (0.33)	1.000 (0.03)	1.000 (0.25)	1.000 (0.13)
F-LSM	5.217			1.000 (0.31)	<0.001 (0.93)	1.000 (0.01)	1.000 (0.27)	1.000 (0.13)
F-LSM	2.363			0.528 (0.43)	0.003 (0.80)	1.000 (0.15)	0.567 (0.42)	1.000 (0.37)
F-MLS	5.373				0.858 (0.39)	1.000 (0.33)	0.007 (0.74)	0.278 (0.48)
F-MLS	2.463				1.000 (0.13)	1.000 (0.27)	1.000 (0.06)	1.000 (0.09)
F-MMM	5.512					0.010 (0.72)	<0.001 (0.99)	<0.001 (0.96)
F-MMM	2.492					1.000 (0.36)	1.000 (0.04)	1.000 (0.24)
F-MSL	5.222						1.000 (0.33)	1.000 (0.15)
F-MSL	2.395						1.000 (0.41)	0.629 (0.18)
F-SLM	5.073							1.000 (0.20)
F-SML	2.480							1.000 (0.15)
	5.155
	2.438

Note. The results are with Bonferroni correction. The values in parentheses are Cohen’s d. Significant results are in boldface.

The interaction between 5Eye and 3Forehead (p = .117, $η_{p}^{2}$ = .045) and the three-way interaction of 5Eye, 3Forehead, and feature attractiveness did not reach significance (p = .060, $η_{p}^{2}$ = .056), indicating that the effect of 5Eye and that of 3Forehead on facial attractiveness was independent, consistent with Experiment 1 and Pallett et al. (2010).

General Discussion

In this study, we manipulated the spatial layouts of facial features from vertical (3Forehead) and horizontal (5Eye) dimensions and adopted a two-alternative forced-choice task in Experiment 1 and a rating task in Experiment 2 to investigate the influence of configural information on facial attractiveness.

First, we found horizontal configural information affected facial attractiveness, which was modulated by feature attractiveness. Both tasks showed that faces with highly attractive features that met the 5Eye criteria were assessed as most attractive, consisting of our hypothesis. However, E-farther was also considered most attractive in both tasks, suggesting that widely spaced eyes are an attractive characteristic, consistent with previous research (Meerdink et al., 1990). However, for faces with low attractive features, a wide eye distance did not offer an attractive advantage; even a narrow eye distance was perceived as better looking in the forced-choice task. The latter may be due to the possibility that the bridge of most low attractive noses is flat, which may be less obvious if the eyes are closer.

Furthermore, in both tasks, faces that deviated too much from the standard 5Eye (i.e., E-furthest and E-closest) were considered unattractive, both for faces with high and low attractive features. This phenomenon could be explained from an evolutionary perspective. The 5Eye distribution might be a general indicator of health, and an abnormal range of eye spacing may indicate some diseases. For example, wide eye distance is a facial feature of some diseases, such as Down syndrome (e.g., Pueschel,1984) and Waardenburg syndrome (e.g., Vichare & Bhargava, 2013). Poor health predicts less attractive (e.g., Rhodes et al., 2007; Weeden & Sabini, 2005).

Second, we found vertical configural information affected facial attractiveness, which was modulated by feature attractiveness. Both tasks showed that those who met the 3Forehead criteria and the faces with short chin (i.e., F-LMS, F-MLS) were judged as most attractive for faces with highly attractive features. A short chin is characteristic of baby face (Berry & McArthur, 1985; Jones & Hill, 1993; McArthur & Apatow, 1984; Zebrowitz & Montepare, 1992) and feminine (Rhodes, 2006), which both are considered attractive to female faces (e.g., Bressan et al., 2009; Jones & Hill, 1993; McArthur & Apatow, 1984; Rhodes, 2006). Besides a small chin, a larger forehead is also considered feminine (Rhodes, 2006) and babyish (Alley, 1981; Berry & McArthur, 1985; Glocker et al., 2009; McArthur & Apatow, 1984; Zebrowitz & Montepare, 1992). These may explain why F-LMS is regarded as more attractive than F-LSM. However, for faces with low attractive features, the effect of “3Forehead” was much smaller, consistent with our hypothesis that configural information’s effect was more significant for faces with highly attractive features. We only found that in the forced-choice task, F-MMM was significantly more attractive than F-SLM, and in the rating task, F-MMM was significantly more attractive than F-LSM.

The larger effects of 5Eye and that of 3Forehead when facial features were highly attractive than when they were low attractive were consistent with Leder et al. (2017) and Ying et al. (2020) in that high attractive faces, compared with low attractive faces, may be more sensitive to holistic processing. These results suggest that the mechanism for perceiving unattractive faces might be different from that for attractive faces. This speculation is worth further investigation.

Finally, the present results are partially consistent with those of Pallett et al. (2010). The null interaction of 3Forehead and 5Eye in both Experiments indicated that the effect of 3Forehead and that of 5Eye were independent, consistent with Pallett et al. (2010). To compare whether 3F5E in the present study fit with the optimal 0.36 length ratio and the optimal 0.46 width ratio in Pallett et al. (2010), we measured the proportion of the most attractive face in Pallett’s study using the same fixed-point as in the present study (see Figure 5).

Figure 5.

The horizontal (top) and vertical (bottom) configuration information of the most attractive face in Pallett et al. (2010; the colorful faces in the most left column) and those in the present study (the grayscale faces in the right three columns).

For the horizontal distribution of features, both studies’ optimal ratios were highly consistent (see Figure 5 top). Most similar to E-standard in the present study, the eyes’ distance of the optimal face in Pallett et al. (2010) is also equal to one eye’s length. However, since the faces in Pallett et al. (2010) did not include the ears, we could not measure the bilateral 1/5 of the 5Eye.

Importantly, the two studies’ optimal vertical ratios were different (see Figure 5 bottom). For faces that meet with 3Forehead criteria in our study, from forehead hairline to eyebrow bone, from eyebrow bone to nose base, from nose base to the bottom of the chin, each takes 1/3 (i.e., 33.33%) length of the face height. Besides, from nose base to edges of underlip and chin takes 1/6 (i.e., 16.67%) length of the face height. Meanwhile, F-LMS and F-MLS were also considered very attractive. Both have a smaller chin (from nose base to the bottom of chin takes 31.82% length of the face height), while F-LMS has a larger forehead that takes 34.85%, and F-MLS has a larger eyebrow to nose distance that takes 34.85%. Nevertheless, in Pallett et al. (2010), from forehead hairline to eyebrow bone took 31.1% length of the face height, while from eyebrow bone to nose base, the ratio was 31.5%, from nose base to the bottom of the chin it was 37.4%. More specifically, from nose base to edges of underlip takes 20.4% length of the face height. Comparing the two studies, we found that our study participants preferred faces with smaller chins to those in Pallett et al. (2010). There were several possible explanations for this discrepancy. First, there might be a cultural difference in optimal facial attractiveness perception. The faces in Pallett et al. (2010) were Caucasian females. The participants were college students in the United States (the majority race supposed to be Caucasian). However, the faces were Chinese females, and the participants were Chinese college students in the present study. The 3F5E may only apply to Chinese participants’ perceiving Chinese faces. While for Caucasian faces, the Marquardt Golden Ratio (Phi) mask (Marquardt, 2002) may apply (Bashour, 2006). The ideal Marquardt mask was found not universal (Hanihara, 2000). Second, the two studies’ manipulation of feature spacing, especially in vertical configurations, was somewhat different. Specifically, we moved the face’s eyes, nose, and mouth vertically and kept the distance from nose base to underlip edges, always taking 1/2 of the distance from nose base to chin’s bottom. However, Pallett et al. (2010) always kept the nose in the same position of a face, moving only the eyes and mouth lengthwise.

Conclusion and Limitations

In conclusion, maintaining the facial features and contours unchanged, just altering the spatial distribution of the facial features, affects facial attractiveness, indicating that configural information impacts facial attractiveness. It was more pronounced for highly attractive faces than low attractive ones. These results extend the role of holistic processing in facial attractiveness perception, provide experimental evidence for the traditional Chinese aesthetic theory—Three Forehead and Five Eyes, and add a new aesthetics rule to existing facial attractiveness literature.

The present study is characterized by some limitations, suggesting avenues for future research. The above conclusion may be limited to the static front faces that we have used in the present study. Face configuration changes across different images of a face (e.g., Burton et al., 2015; Noyes & Jenkins, 2017), which show different attractiveness (Jenkins et al., 2011; White et al., 2017). Whether the changing configuration could account for the different attractiveness of the same face’s different images is worth investigating. Another limitation of the present study is that the faces in the present study were all Chinese female faces. The above conclusion might not be generalized to male faces, although we did not find the main effect of participants’ gender nor its interactions with other variables. It is also an open question whether Chinese/Western participants perceive the attractiveness of own-race faces and other-race faces similarly if the same configural manipulation was applied to them. Research had found that when recognizing faces, the Japanese are more configural than Americans (Miyamoto et al., 2011). Future research could also focus on whether people in different cultures have different sensitivities on facial configurations when evaluating facial attractiveness.

Footnotes

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This research was supported by the grant from the National Natural Science Foundation of China (32071048, 31771208) to Guomei Zhou.

ORCID iD

Guomei Zhou

References

Abbas

Z.-A.

Duchaine

(2008). The role of holistic processing in judgments of facial attractiveness. Perception, 37, 1187–1196. https://doi.org/10.1068/p5984

Alley

(1981). Head shape and the perception of cuteness. Developmental Psychology, 17, 650–654. https://doi.org/10.1037/0012-1649.17.5.650

Bashour

(2006). An objective system for measuring facial attractiveness. Plastic and Reconstructive Surgery, 118, 757–776. https://doi.org/10.1097/01.prs.0000207382.60636.1c

Baudouin

J.-Y.

Humphreys

G. W.

(2006). Configural information in gender categorisation. Perception, 35, 531–540. https://doi.org/10.1068/p3403

Berry

D. S.

McArthur

L. Z.

(1985). Some components and consequences of a babyface. Journal of Personality and Social Psychology, 48, 312–323. https://doi.org/10.1037/0022-3514.48.2.312

Bressan

Bertamini

Nalli

Zanutto

(2009). Men do not have a stronger preference than women for self-resemblant child faces. Archives of Sexual Behavior, 38, 657–664. https://doi.org/10.1007/s10508-008-9350-0

Burke

Sulikowski

(2013). The evolution of holistic processing of faces. Frontiers in Psychology, 4, 11. https://doi.org/10.3389/fpsyg.2013.00011

Burton

A. M.

Schweinberger

S. R.

Jenkins

Kaufmann

J. M.

(2015). Arguments against a configural processing account of familiar face recognition. Perspectives on Psychological Science, 10, 482–496.

Calder

A. J.

Young

A. W.

Keane

Dean

(2000). Configural information in facial expression perception. Journal of Experimental Psychology: Human Perception and Performance, 26, 527–551. https://doi.org/10.1037//0096-1523.26.2.527

10.

Chang

Chou

C.-H.

(2009). A bi-prototype theory of facial attractiveness. Neural Computation, 21, 890–910. https://doi.org/10.1162/neco.2008.07-07-566

11.

Collishaw

S. M.

Hole

G. J.

(2000). Featural and configurational processes in the recognition of faces of different familiarity. Perception, 29, 893–909.

12.

Glocker

M. L.

Langleben

D. D.

Ruparel

Loughead

J. W.

Gur

R. C.

Sachser

(2009). Baby schema in infant faces induces cuteness perception and motivation for caretaking in adults. Ethology, 115, 257–263. https://doi.org/10.1111/j.1439-0310.2008.01603.x

13.

Goffaux

Hault

Michel

Vuong

Q. C.

Rossion

(2005). The respective role of low and high spatial frequencies in supporting configural and featural processing of faces. Perception, 34, 77–86.

14.

Goffaux

Rossion

(2006). “ Faces are spatial”—Holistic face perception is supported by low spatial frequencies. Journal of Experimental Psychology: Human Perception and Performance, 32, 1023–1039.

15.

Hancock

P. J. B.

Bruce

Miyamoto

A. M.

(2000). Recognition of unfamiliar faces. Trends in Cognitive Sciences, 4, 330–337. https://doi.org/10.1016/S1364-6613(00)01519-9

16.

Hanihara

(2000). Frontal and facial flatness of major human populations. American Journal of Physical Anthropology, 111, 105–134.

17.

Hayward

W. G.

Crookes

Rhodes

(2013). The other-race effect: Holistic coding differences and beyond. Visual Cognition, 21, 1224–1247.

18.

Jenkins

White

Van Montfort

Burton

A. M.

(2011). Variability in photos of the same face. Cognition, 121, 313–323.

19.

Jones

Hill

(1993). Criteria of facial attractiveness in five populations. Human Nature, 4, 271–296. https://doi.org/10.1007/BF02692202

20.

Leder

Goller

Forster

Schlageter

Paul

M. A.

(2017). Face inversion increases attractiveness. Acta Psychologica, 178, 25–31. https://doi.org/10.1016/j.actpsy.2017.05.005

21.

Little

A. C.

Jones

B. C.

DeBruine

L. M.

Feinberg

D. R.

(2008). Symmetry and sexual dimorphism in human faces: Interrelated preferences suggest both signal quality. Behavioral Ecology, 19, 902–908. https://doi.org/10.1093/beheco/arn049

22.

Liu

Chen

(2018). The boundary of holistic processing in the appraisal of facial attractiveness. Royal Society Open Science, 5, 171616. https://doi.org/10.1098/rsos.171616

23.

Liu, X. (2020). Group Size Effect and its neural correlates in attractiveness perception: Take faces and flowers for examples. [Doctoral dissertation. Sun Yat-sen University].

24.

Marquardt

S. R.

(2002). Dr. Stephen R. Marquardt on the Golden Decagon and human facial beauty. Interview by Dr. Gottlieb. Journal of Clinical Orthodontics, 36, 339–347.

25.

Maurer

Grand

R. L.

Mondloch

C. J.

(2002). The many faces of configural processing. Trends in Cognitive Sciences, 6, 255–260.

26.

McArthur

L. Z.

Apatow

(1984). Impressions of baby-faced adults. Social Cognition, 2, 315–342. https://doi.org/10.1521/soco.1984.2.4.315

27.

McKone, E., Davies, A. A., Darke, H., Crookes, K., Wickramariyaratne, T., Zappia, S., Fiorentini, C., Favelle, S., Broughton, M., & Fernando, D. (2013). Importance of the inverted control in measuring holistic face processing with the composite effect and part-whole effect. Frontiers in Psychology, 4, 33. https://doi.org/10.3389/fpsyg.2013.00033

28.

Meerdink

J. E.

Garbin

C. P.

Leger

D. W.

(1990). Cross-gender perceptions of facial attributes and their relation to attractiveness: Do we see them differently than they see us? Perception & Psychophysics, 48, 227–233. https://doi.org/10.3758/BF03211522

29.

Michel

Corneille

Rossion

(2007). Race categorization modulates holistic face encoding. Cognitive Science: A Multidisciplinary Journal, 31, 911–924. https://doi.org/10.1080/03640210701530805

30.

Miyamoto

Yoshikawa

Kitayama

(2011). Feature and configuration in face processing: Japanese are more configural than Americans. Cognitive Science, 35, 563–574. https://doi.org/10.1111/j.1551-6709.2010.01163.x

31.

Noyes

Jenkins

(2017). Camera-to-subject distance affects face configuration and perceived identity. Cognition, 165, 97–104.

32.

Pallett

P. M.

Link

Lee

(2010). New “golden” ratios for facial beauty. Vision Research, 50, 149–154. https://doi.org/10.1016/j.visres.2009.11.003

33.

Pueschel

S. M.

(1984). The young child with Down syndrome. Human Sciences Press.

34.

Rhodes

(2006). The evolutionary psychology of facial beauty. Annual Review of Psychology, 57, 199–226. https://doi.org/10.1146/annurev.psych.57.102904.190208

35.

Rhodes

Yoshikawa

Palermo

Simmons

L. W.

Peters

Lee

, … Crawford

J. R.

(2007). Perceived health contributes to the attractiveness of facial symmetry, averageness, and sexual dimorphism. Perception, 36, 1244–1252. https://doi.org/10.1068/p5712

36.

Richler

J. J.

Gauthier

(2014). A meta-analysis and review of holistic face processing. Psychological Bulletin, 140, 1281–1302. https://doi.org/10.1037/a0037004

37.

Rossion

(2013). The composite face illusion: A whole window into our understanding of holistic face perception. Visual Cognition, 21, 139–253.

38.

Stróżak

Zielińska

(2019). Different processes in attractiveness assessments for unattractive and highly attractive faces—The role of presentation duration and rotation. Acta Psychologica, 200, 102946. https://doi.org/10.1016/j.actpsy.2019.102946

39.

Vichare

Bhargava

(2013). Waardenburg syndrome: A rare case with bilateral congenital cataract: An unusual entity. Medical Journal Armed Forces India, 69, 172–174.

40.

Weeden

Sabini

(2005). Physical attractiveness and health in western societies: A review. Psychological Bulletin, 131, 635–653. https://doi.org/10.1037/0033-2909.131.5.635

41.

White

Sutherland

C. A.

Burton

A. L.

(2017). Choosing face: The curse of self in profile image selection. Cognitive Research: Principles and Implications, 2, 1–9.

42.

Yin

R. K.

(1969). Looking at upside-down faces. Journal of Experimental Psychology, 81, 141–145. https://doi.org/10.1037/h0027474

43.

Ying

Cheng

Deepu Rajan

(2020). Perception of attractive and unattractive face groups is driven by distinct spatial frequencies. PsyCh Journal, 9, 804–818. https://doi.org/10.1002/pchj.386

44.

Yu, J. (1957). Classified compilation of ancient Chinese painting. Beijing: People's Fine Arts Publishing House.

45.

Zebrowitz

L. A.

Montepare

J. M.

(1992). Impressions of babyfaced individuals across the life span. Developmental Psychology, 28, 1143–1152.

46.

Zhang

H. K.

Wang

Luo

Zheng

Z. J.

Jiang

(2017). A study of the best fixed-point of facial proportions. Medicine & Philosophy, (B), 38, 64–69.

47.

Zhao

Hayward

W. G.

(2010). Holistic processing underlies gender judgments of faces. Attention, Perception, & Psychophysics, 72, 591–596.