From Calibrated Morphs to Facial Stimuli: The Beauty of a Statistically Informed Picture

The interest in physical appearance and attractiveness is presumably much older than modern humans. The depiction of the human face and body is the subject of many artworks including the Upper Paleolithic Venus figurines (e.g., the Venus of Willendorf, c. 30 000 years ago) and many paintings and sculptures of the ancient Egyptians, Greeks, and Romans. The scientific inquiry into the systematic variation of physical appearance also has a long history. Egyptian artisans already used square grids and standard proportions to produce consistent depictions of human and other figures (Robins 1994). In “Modern Morphometrics in Physical Anthropology,” Dennis E. Slice (2005, 1) summarized more recent developments in the scientific inquiry of the human physique as follows: “Johann Sigismund Elsholtz formalized the scientific measurement of living individuals, anthropometry, in his 1654 Doctoral dissertation (Kolar and Salter 1996), and his particular interest in symmetry would appeal to many present-day anthropologists and general biologists. From the 19th century to the present day, the measurement and analysis of human beings and their skeletal remains have been a central theme in anthropology, though not always with beneficent motivation (e.g., Gould 1981). During this time, anthropologists have often taken advantage of the state-of-the-art in statistical methodology, but they have not been just passive consumers of technological innovation. Indeed, pervasive interest in our own species, its artifacts, and our closest relatives has motivated and contributed much to the development of statistical methods that are now taken for granted in areas far afield from anthropology. The early work of the biometric laboratory established by Galton and Pearson bears witness to the vital interplay between the development of statistical methodology and anthropological research (e.g., Mahalanobis 1928, 1930; Morant 1928, 1939; Pearson 1903, 1933).”

The dynamic interplay between physical anthropology and statistical advancements persists because emerging methods in shape analysis are often driven by anthropological inquiries. Conversely, the introduction of novel morphometric tools fosters new research opportunities and presents robust alternatives to conventional approaches. Key contributions encompass projections of future directions in morphometrics, advancements in shape analysis methodologies, and examples illustrating how state-of-the-art morphometric techniques help address fundamental research questions.

By about 1880, Francis Galton combined photographic portraits into composite images leading to the observation that similar features of family members were particularly defined (Galton 1878). More than 100 years later, the technique was resurrected and facilitated empirical approaches to understanding human responses to facial variation. Some scholars will remember the video of Michael Jackson's song “Black or White” in 1991, which showed facial transformations of members from different ethnic groups. The technique of “feature-based image metamorphosis” was described a year later by Beier and Neely (1992) and impacted subsequent biological and psychological inquiries designed to identify the motives for human assessments of feature-based facial variation. Facial morphing was adopted to investigate the attractiveness of facial composites from different ethnic origins (Grammer and Thornhill 1994; Perrett et al. 1994)—later it was extended to the study of facial sexual dimorphism (Perrett et al. 1998) and, more generally, to research questions regarding the relationships between systematic variation in facial images and social attribution.

A morphing operation is defined as the process that changes one image into another through a seamless transition (see Aloraibi 2023 for review). Traditional face morphing uses one image as the source and a second as the destination. A correspondence between the two is established via pairs of somatometric feature primitives such as curves, mesh nodes, line segments, and points of interest that can be reliably detected. The image morphing algorithm often maps coordinates between two (or more) digital images based on several such pre-defined landmarks (e.g., the inner and outer corners of the eyes). These coordinates are used to “warp” and cross-dissolve images based on corresponding landmarks. Facial images are delineated using landmarks that provide triangle-to-triangle correspondences for mesh-warping. Thus, morphing one face into another first transforms the shape and then adds texture information to create increments between the source and the destination image (https://normankarr.com/computational-photography/face-morphing/ for illustrative visualization of the steps).

The morphing technique supplied researchers with a fascinating tool for addressing questions about the associations between facial appearance and perception. For example, scholars created digital composites from the portraits of participants known to differ in behavioral traits and had them rated for social attributes (e.g., Boothroyd et al. 2008). Between 1990 and 2010, this and similar approaches to creating realistic-looking faces using image morphing were the gold standard in face research. The methodology has been improved and refined over the years, facilitated by developments in computer graphics. This was certainly an advancement over previous approaches in face research confined to single-image presentation or the creation of charismatic faces using line drawings (interestingly, recent developments in artificial intelligence show that photorealistic facial images can be created from freehand sketches; Chen et al. 2020). Regardless of the valuable insights into the social perception of faces, limitations are associated with this type of morphing: From a biological perspective, two essential questions remained—(1) Which biological processes do transformation vectors between source and destination images represent [causes of facial shape variation]? (2) Which aspects of natural facial variation inform the perception of a social trait [consequences of facial shape variation]? Addressing biological causation requires a method tied to biological reasoning.

As the proper description and statistical analysis of shape variation within and among samples of organisms, along with the analysis of shape change as a result of growth and evolution, are the basis for testing hypotheses in evolutionary theory and presenting new findings, we highly depend on the power of the morphometric methods used. The approach that commonly had been used in physical anthropology is referred to as traditional morphometrics (Reyment 1991; Marcus 1990) or multivariate morphometrics (Blackith and Reyment 1971). Rohlf and Marcus (1993, 129) stated that traditional morphometrics are “characterized by the application of multivariate statistical methods to sets of variables.” The variables usually correspond to various distance measurements—as well as areas and volumes on an organism, usually captured by calipers as lengths and widths of structures and distances between landmarks. “Sometimes angles and ratios are used. […] The results are mostly expressed numerically and graphically in terms of linear combinations of the measured variables. Examples of the techniques used are principal component analysis, canonical variate analysis, discriminant functions and generalized distances.” The inherent restriction is that the shape of the original form is not recoverable. An investigator may know, for example, that several distance measurements share the same landmark, but this information is not included in the multivariate analysis. Accordingly, the analysis cannot be expected to be as powerful as if all information was included. Geometric morphometric methods (GMM) do exactly this. The data recording via Cartesian coordinates captures and preserves the geometry of the structure being studied throughout the statistical analyses (Bookstein 1991). Moreover, the results of the statistical analyses can be visualized again as forms, which are much more readily interpretable than large tables of numbers alone.

By the mid 2000s, GMM had been increasingly adopted and advanced in physical anthropology and the study of human evolution. At that time, two of us (BF and KS) published geometric morphometric analyses mapping biological measures (such as testosterone, Fink et al. 2005) as well as first impressions (such as attractiveness, Schaefer et al. 2006) onto facial shape. Schaefer et al. (2009) formalized the comparison of biological causes of facial shape variation and their impact on social perception in the concept of the Psychomorphospace. The authors showed that GMM enabled identifying facial shapes and features that mediate the relationship between physical measures and face perception.

In addition to these merits, calibrated morphs are valuable for evaluating sensitive topics and vulnerable groups, such as stereotyping and stigmatization, within and across populations by protecting participant identity.

Within an evolutionary framework, our (2011) article in the American Journal of Human Biology established that male physical strength and facial attractiveness are not the same construct in members of a WEIRD (Western Educated Industrialized Rich Democratic, Henrich et al. 2010) society. We decomposed the association of male facial shape with physical formidability (Sell et al. 2009; muscular strength, body height, body fat, shoulder width) and appearance (attractiveness/dominance/masculinity ratings). For each of these variables, facial shape changes were plotted as photorealistic portraits for the first time (see color figure in our original article). The calibrated morphs have bridged the gap between statistical models of morphological patterns and their intuitive interpretation, making the methods and findings accessible to a broader audience. Figure 1 displays a word cloud derived from the titles of the articles citing Windhager et al. (2011). The multidisciplinary scholarly reception of the article is illustrated by bibliometric data retrieved from the Scopus database. A ranked list of Scopus subject areas shows the distribution across disciplines. The largest share of citations stems from the field of Psychology (24%), followed by Agricultural and Biological Sciences (13%), Medicine (11%), and Biochemistry, Genetics and Molecular Biology (11%). Further subject areas include Social Sciences (10%), Neuroscience (8%), Arts and Humanities (7%), Multidisciplinary (7%), Computer Science, Engineering, and other fields (9%). The temporal distribution of citations reflects sustained interest.

Challenges for future work include integrating the extensive knowledge from fields such as craniofacial biology and orthodontics to predict facial appearance. Previous research documented, for example, morphological covariation between the basicranium and the viscerocranium (i.e., short-faced versus long-faced head forms; Bhat and Enlow 1985) within and across human groups (e.g., Bastir and Rosas 2006), (population) genetic effects (e.g., Hallgrimsson et al. 2019; White et al. 2021), and interactions with environmental factors such as nutrition (e.g., von Cramon-Taubadel 2011) and bioclimate (e.g., Matsumura et al. 2024). Furthermore, the interplay between hard and soft tissues (e.g., Zedníková Malá et al. 2018) deserves closer scrutiny. This holistic human biological perspective may offer a deeper understanding of the association between facial bony structures and soft facial features.

The American Journal of Human Biology made the “beauty” of calibrated morphs accessible to an interdisciplinary audience. This approach allows quantitatively assessing and simulating forms to test a wide range of theories and hypotheses within and beyond the evolutionary context. It even extends to artificial variations in form (e.g., car fronts or houses; Windhager et al. 2012). On the celebration of its 50th anniversary, we express our appreciation to the Human Biological Association for fostering the understanding of human biological variation.

The authors declare no conflicts of interest.