Facial Symmetry: What the Research Says and What a Mesh Can Measure
Symmetry is the most discussed number in facial analysis, and the most over-claimed.
What the literature suggests
Rhodes (2006) reviewed the evidence and reported, from a critical review and meta-analyses, that averageness, symmetry and sexual dimorphism are attractive in both male and female faces and across cultures. Little, Jones and DeBruine (2011) review symmetry among several characteristics that influence attractiveness judgements. Fink, Neave and Manning (2006) asked college students to rate 20 female faces that varied in symmetry, and report higher ratings of attractiveness, health and some personality traits for more symmetric faces. So the literature does link symmetry to perceived attractiveness. But those studies manipulate or measure symmetry under controlled conditions, often with small stimulus sets, and they study how raters respond. They do not validate a symmetry number computed from a single casual selfie, and they do not tell you what your number means for you.
What Selfbaus computes
The engine takes a set of paired landmarks that sit on opposite sides of the face, such as outer eye corners, inner eye corners, brow points, mouth corners and points along the face oval. After rotating the landmarks so the eye line is level, it reflects one side across the face's midline and measures how far each mirrored point lands from its partner, then averages the displacements, expressed in iris-distance units. Version 0.2.0 uses eight mirrored pairs: outer eye corners (mesh points 33 and 263), inner eye corners (133, 362), two brow pairs (70, 300 and 105, 334), mouth corners (61, 291), and three face-oval pairs (234, 454; 172, 397; 58, 288). A value near zero means the mesh points line up when mirrored.
What that number includes
- Real asymmetry of the face.
- Head turn and tilt, even small ones.
- Uneven expression, like a half smile.
- Hair, shadows and glasses that move landmarks.
- Mesh error on oval points, which are estimated rather than seen.
Because of these, we describe the output as capture-conditioned mesh alignment. It is not a medical or aesthetic grade. Everyone has some asymmetry.
How to use it
Take a few frontal, neutral-expression photos in the same light. If the number moves between shots more than between faces, you are reading capture noise.
References
- Rhodes, G. (2006). The evolutionary psychology of facial beauty. Annual Review of Psychology. https://doi.org/10.1146/annurev.psych.57.102904.190208
- Little, A. C., Jones, B. C., & DeBruine, L. M. (2011). Facial attractiveness: evolutionary based research. Phil. Trans. R. Soc. B. https://doi.org/10.1098/rstb.2010.0404
- Fink, B., Neave, N., & Manning, J. T. (2006). Facial symmetry and judgements of attractiveness, health and personality. Personality and Individual Differences. https://doi.org/10.1016/j.paid.2006.01.017