The Biology of Facial Symmetry
Perfect bilateral symmetry is the theoretical developmental ideal, genes code for a symmetric face, and developmental noise during growth introduces asymmetries. Facial symmetry therefore serves as a signal of developmental stability: low asymmetry indicates that development proceeded smoothly without disruption from genetic or environmental stressors. This is why symmetry is associated with perceived attractiveness and health across cultures.
In reality, no human face is perfectly symmetric. Even highly symmetric-looking faces, when measured precisely, show asymmetries in eye height, nose orientation, and jaw position. The average face asymmetry is small but measurable, and the distribution has a long tail, some individuals have quite pronounced asymmetries resulting from developmental or medical factors.
Does Symmetry Improve Face Recognition Accuracy?
For deep learning face recognition systems, symmetry has a modest positive effect. Highly symmetric faces produce more stable embeddings across different photo conditions because the distribution of facial features is more consistent from slightly different angles. A slightly off-centre photo of a symmetric face will show similar feature positions to a straight-on photo; for an asymmetric face, the shift can be more pronounced.
However, symmetry is far from the dominant factor in recognition accuracy. Image quality, lighting consistency, and face alignment quality each have larger effects on embedding stability. A highly asymmetric face photographed well will produce more accurate results than a symmetric face photographed under poor conditions.