What Is Facial Geometry?

Facial recognition systems don't store images of faces — they work with mathematical representations of facial geometry. Landmark detection algorithms identify the positions of key facial features (typically 68-468 points) in an image. Geometry measurements (distances, ratios, angles between these points) are then extracted and encoded into a face embedding. These geometric measurements are remarkably stable across different photographs of the same person: lighting, makeup, aging, and even minor surgery rarely change the fundamental geometric relationships enough to prevent matching. This stability is what makes biometric face scanning effective for NCII detection across differently-encoded or AI-generated versions of a victim's likeness.

Key facts about this term

1. Facial geometry is unique and stable across images The geometric relationships between facial landmarks are as unique as fingerprints and remain consistent across variations in lighting, age, makeup, and image quality.
2. Facial geometry is preserved in AI-generated deepfakes When AI generates a realistic face swap, it must preserve the target person's facial geometry to produce a convincing likeness. This is why biometric face matching detects deepfakes even when pixel-level hashing would miss them.
3. Geometric measurements survive image re-encoding Re-encoding an image (changing resolution, format, or compression) changes pixels but not the underlying facial geometry. Biometric systems extract the same geometric measurements regardless of encoding.