Figure 5 - available via license: Creative Commons Attribution 4.0 International
Content may be subject to copyright.
Expression/antiexpression dissimilarity, collapsing over identity. Matrices show ground truth dissimilarity for motion metrics max (A), slope (B), and mid (C) with multidimensional scaling (MDS) plots for max (D), slope (E), and mid (F). Other matrices show perceived dissimilarity for Studies 1 (G) and 2 (H), with MDS plots for Studies 1 (I) and 2 (J). Caricature levels of rows and columns in dissimilarity matrices and of letters in MDS plots are coloured as in the legend. Caricatures are shown in bold. A = anger, D = disgust, F = fear, H = happy, S = surprise. Caricatures appear more spread out in space compared to anticaricatures.
Source publication
Although faces “in the wild” constantly undergo complicated movements, humans adeptly perceive facial identity and expression. Previous studies, focusing mainly on identity, used photographic caricature to show that distinctive form increases perceived dissimilarity. We tested whether distinctive facial movements showed similar effects, and we focu...
Contexts in source publication
Context 1
... first step involved computing the average dissimilarity between all pairs of expressions/antiexpressions (collapsing over identities). We display in Figure 5 these recomputed dissimilarity matrices with 4 caricature levels × 5 basic expressions = 20 rows and columns for motion metrics max ( Figure 5A), slope (Figure 5B), mid ( Figure 5C), and the same matrices for participants' perceived dissimilarity for Study 1 ( Figure 5G) and Study 2 ( Figure 5H). In the second step, we projected these expression category-specific dissimilarities onto a 2-dimensional plane using the best-fitting metricstress MDS solutions from 5000 random starting values (mdscale.m in MATLAB R2015B, The Mathworks, Nattick, MA) for motion metrics max ( Figure 5D), slope ( Figure 5E), mid ( Figure 5F) and participants' perceived dissimilarity for Study 1 ( Figure 5I) and Study 2 ( Figure 5J). ...
Context 2
... first step involved computing the average dissimilarity between all pairs of expressions/antiexpressions (collapsing over identities). We display in Figure 5 these recomputed dissimilarity matrices with 4 caricature levels × 5 basic expressions = 20 rows and columns for motion metrics max ( Figure 5A), slope (Figure 5B), mid ( Figure 5C), and the same matrices for participants' perceived dissimilarity for Study 1 ( Figure 5G) and Study 2 ( Figure 5H). In the second step, we projected these expression category-specific dissimilarities onto a 2-dimensional plane using the best-fitting metricstress MDS solutions from 5000 random starting values (mdscale.m in MATLAB R2015B, The Mathworks, Nattick, MA) for motion metrics max ( Figure 5D), slope ( Figure 5E), mid ( Figure 5F) and participants' perceived dissimilarity for Study 1 ( Figure 5I) and Study 2 ( Figure 5J). ...
Context 3
... first step involved computing the average dissimilarity between all pairs of expressions/antiexpressions (collapsing over identities). We display in Figure 5 these recomputed dissimilarity matrices with 4 caricature levels × 5 basic expressions = 20 rows and columns for motion metrics max ( Figure 5A), slope (Figure 5B), mid ( Figure 5C), and the same matrices for participants' perceived dissimilarity for Study 1 ( Figure 5G) and Study 2 ( Figure 5H). In the second step, we projected these expression category-specific dissimilarities onto a 2-dimensional plane using the best-fitting metricstress MDS solutions from 5000 random starting values (mdscale.m in MATLAB R2015B, The Mathworks, Nattick, MA) for motion metrics max ( Figure 5D), slope ( Figure 5E), mid ( Figure 5F) and participants' perceived dissimilarity for Study 1 ( Figure 5I) and Study 2 ( Figure 5J). ...
Context 4
... first step involved computing the average dissimilarity between all pairs of expressions/antiexpressions (collapsing over identities). We display in Figure 5 these recomputed dissimilarity matrices with 4 caricature levels × 5 basic expressions = 20 rows and columns for motion metrics max ( Figure 5A), slope (Figure 5B), mid ( Figure 5C), and the same matrices for participants' perceived dissimilarity for Study 1 ( Figure 5G) and Study 2 ( Figure 5H). In the second step, we projected these expression category-specific dissimilarities onto a 2-dimensional plane using the best-fitting metricstress MDS solutions from 5000 random starting values (mdscale.m in MATLAB R2015B, The Mathworks, Nattick, MA) for motion metrics max ( Figure 5D), slope ( Figure 5E), mid ( Figure 5F) and participants' perceived dissimilarity for Study 1 ( Figure 5I) and Study 2 ( Figure 5J). ...
Context 5
... first step involved computing the average dissimilarity between all pairs of expressions/antiexpressions (collapsing over identities). We display in Figure 5 these recomputed dissimilarity matrices with 4 caricature levels × 5 basic expressions = 20 rows and columns for motion metrics max ( Figure 5A), slope (Figure 5B), mid ( Figure 5C), and the same matrices for participants' perceived dissimilarity for Study 1 ( Figure 5G) and Study 2 ( Figure 5H). In the second step, we projected these expression category-specific dissimilarities onto a 2-dimensional plane using the best-fitting metricstress MDS solutions from 5000 random starting values (mdscale.m in MATLAB R2015B, The Mathworks, Nattick, MA) for motion metrics max ( Figure 5D), slope ( Figure 5E), mid ( Figure 5F) and participants' perceived dissimilarity for Study 1 ( Figure 5I) and Study 2 ( Figure 5J). ...
Context 6
... display in Figure 5 these recomputed dissimilarity matrices with 4 caricature levels × 5 basic expressions = 20 rows and columns for motion metrics max ( Figure 5A), slope (Figure 5B), mid ( Figure 5C), and the same matrices for participants' perceived dissimilarity for Study 1 ( Figure 5G) and Study 2 ( Figure 5H). In the second step, we projected these expression category-specific dissimilarities onto a 2-dimensional plane using the best-fitting metricstress MDS solutions from 5000 random starting values (mdscale.m in MATLAB R2015B, The Mathworks, Nattick, MA) for motion metrics max ( Figure 5D), slope ( Figure 5E), mid ( Figure 5F) and participants' perceived dissimilarity for Study 1 ( Figure 5I) and Study 2 ( Figure 5J). An MDS analysis of expression categories using Study 1 data was previously reported in Furl et al. (2020). ...
Context 7
... display in Figure 5 these recomputed dissimilarity matrices with 4 caricature levels × 5 basic expressions = 20 rows and columns for motion metrics max ( Figure 5A), slope (Figure 5B), mid ( Figure 5C), and the same matrices for participants' perceived dissimilarity for Study 1 ( Figure 5G) and Study 2 ( Figure 5H). In the second step, we projected these expression category-specific dissimilarities onto a 2-dimensional plane using the best-fitting metricstress MDS solutions from 5000 random starting values (mdscale.m in MATLAB R2015B, The Mathworks, Nattick, MA) for motion metrics max ( Figure 5D), slope ( Figure 5E), mid ( Figure 5F) and participants' perceived dissimilarity for Study 1 ( Figure 5I) and Study 2 ( Figure 5J). An MDS analysis of expression categories using Study 1 data was previously reported in Furl et al. (2020). ...
Context 8
... display in Figure 5 these recomputed dissimilarity matrices with 4 caricature levels × 5 basic expressions = 20 rows and columns for motion metrics max ( Figure 5A), slope (Figure 5B), mid ( Figure 5C), and the same matrices for participants' perceived dissimilarity for Study 1 ( Figure 5G) and Study 2 ( Figure 5H). In the second step, we projected these expression category-specific dissimilarities onto a 2-dimensional plane using the best-fitting metricstress MDS solutions from 5000 random starting values (mdscale.m in MATLAB R2015B, The Mathworks, Nattick, MA) for motion metrics max ( Figure 5D), slope ( Figure 5E), mid ( Figure 5F) and participants' perceived dissimilarity for Study 1 ( Figure 5I) and Study 2 ( Figure 5J). An MDS analysis of expression categories using Study 1 data was previously reported in Furl et al. (2020). ...
Context 9
... display in Figure 5 these recomputed dissimilarity matrices with 4 caricature levels × 5 basic expressions = 20 rows and columns for motion metrics max ( Figure 5A), slope (Figure 5B), mid ( Figure 5C), and the same matrices for participants' perceived dissimilarity for Study 1 ( Figure 5G) and Study 2 ( Figure 5H). In the second step, we projected these expression category-specific dissimilarities onto a 2-dimensional plane using the best-fitting metricstress MDS solutions from 5000 random starting values (mdscale.m in MATLAB R2015B, The Mathworks, Nattick, MA) for motion metrics max ( Figure 5D), slope ( Figure 5E), mid ( Figure 5F) and participants' perceived dissimilarity for Study 1 ( Figure 5I) and Study 2 ( Figure 5J). An MDS analysis of expression categories using Study 1 data was previously reported in Furl et al. (2020). ...
Context 10
... display in Figure 5 these recomputed dissimilarity matrices with 4 caricature levels × 5 basic expressions = 20 rows and columns for motion metrics max ( Figure 5A), slope (Figure 5B), mid ( Figure 5C), and the same matrices for participants' perceived dissimilarity for Study 1 ( Figure 5G) and Study 2 ( Figure 5H). In the second step, we projected these expression category-specific dissimilarities onto a 2-dimensional plane using the best-fitting metricstress MDS solutions from 5000 random starting values (mdscale.m in MATLAB R2015B, The Mathworks, Nattick, MA) for motion metrics max ( Figure 5D), slope ( Figure 5E), mid ( Figure 5F) and participants' perceived dissimilarity for Study 1 ( Figure 5I) and Study 2 ( Figure 5J). An MDS analysis of expression categories using Study 1 data was previously reported in Furl et al. (2020). ...
Context 11
... expressions and antiexpressions (large bold letters) appear more spaced out and surround anticaricatures (smaller, lighter letters), which cluster towards the centres of the plots. This pattern is clearest for slope ( Figure 5E) and, for all other plots, is most prominent for the first MDS dimension (x-axis). This "spacing out" of caricatured expressions provides another view on the caricature effects for "expression-differ" pairs shown in Figure 4D and E. Antiexpressions tend to occupy opposite sides of face space from their corresponding expressions. ...
Similar publications
Faces carry key personal information about individuals, including cues to their identity, social traits, and emotional state. Much research to date has employed static images of faces taken under tightly controlled conditions yet faces in the real world are dynamic and experienced under ambient conditions. A common approach to studying key dimensio...
Citations
... Caricaturing dynamic faces poses a further challenge as the process must account for both spatial and temporal patterns of variation. Previous approaches have manipulated the magnitude of motion in facial landmarks during simple facial behaviours, using the resulting motion vectors to drive virtual head models (Hill et al., 2005;Furl et al., 2020Furl et al., , 2022. However, such artificial head models lack many of the features present in real faces. ...
... Facial caricatures offer an important tool for studying face perception by allowing parametric manipulation of key facial dimensions that would be difficult or impossible for a person to pose naturally. Caricaturing manipulations predict behavioural ratings of corresponding facial features (Benson and Perrett, 1991a;Burt and Perrett, 1995;Calder et al., 1997Calder et al., , 2000Blanz et al., 2000;Furl et al., 2022), produce perceptual adaptation effects (Leopold et al., 2001;Jiang et al., 2006;Juricevic and Webster, 2012), and predict neural responses in face-selective regions (Furl et al., 2020). Our approach extends existing spatial caricaturing techniques by providing a method for generating dynamic anti-caricatures from natural facial behaviours, thereby allowing investigation of both spatial and temporal features underlying face perception. ...
... Such behaviours could, for instance, include poses of emotional expressions or head turns. Dynamic caricatures of facial behaviours could be applied to study recognition of those behaviours (Furl et al., 2020(Furl et al., , 2022, similar to how static caricatures have been used to characterise recognition of facial identity (Leopold et al., 2001). Indeed, our own behavioural results confirmed that observers' perception of facial dynamicity increased with increasing levels of caricaturing. ...
Faces carry key personal information about individuals, including cues to their identity, social traits, and emotional state. Much research to date has employed static images of faces taken under tightly controlled conditions yet faces in the real world are dynamic and experienced under ambient conditions. A common approach to studying key dimensions of facial variation is the use of facial caricatures. However, such techniques have again typically relied on static images, and the few examples of dynamic caricatures have relied on animating graphical head models. Here, we present a principal component analysis (PCA)-based active appearance model for capturing patterns of spatiotemporal variation in videos of natural dynamic facial behaviours. We demonstrate how this technique can be applied to generate dynamic anti-caricatures of biological motion patterns in facial behaviours. This technique could be extended to caricaturing other facial dimensions, or to more general analyses of spatiotemporal variations in dynamic faces.
