AI Face Shape Detection Guide & Geometric Methodology
An authoritative technical guide explaining how modern computer vision, 468 three-dimensional facial landmarks, and classical Farkas anthropometry objectively classify human facial morphology into seven distinct geometric archetypes.
Craniofacial biometric classification has evolved from manual caliper measurements into real-time computer vision photogrammetry. While historical profiling relied on subjective appraisals, contemporary artificial intelligence identifies geometric structures through dense 3D coordinate meshes. This guide details the clinical photogrammetry, morphological indices, and algorithmic pipelines used to classify the human face into seven classical archetypes directly within a web browser.
How AI Detects Your Face Shape from a Single Photo
Automated face shape recognition operates on dense coordinate topologies derived from neural regression networks. The detector registers 468 distinct anatomical coordinates across the facial surface in three dimensions (X, Y, and Z depth planes). Rather than using arbitrary contour silhouettes, the system applies clinical craniofacial anthropometry established by Dr. Leslie Farkas to determine invariant geometric ratios.
Primary Anatomical Landmarks in Craniofacial Geometry
- tr (Trichion):Midpoint of the superior hairline boundary defining the top of the upper cranial third.
- ft (Frontotemporale):Lateral points on each temple defining minimal frontal breadth.
- zy (Zygion):Most lateral margins of the zygomatic arches establishing bizygomatic breadth.
- go (Gonion):Posterior angles of the mandible, quantifying bigonial jaw width.
- n (Nasion) & sn (Subnasale):Reference points demarcating the nasal bridge and the midface boundary.
- gn / me (Gnathion / Menton):Lowest anatomical points defining the bony chin synthesis.
Following 3D coordinate registration and perspective rectification to counteract pitch, yaw, and roll, four mathematical ratio vectors are computed:
1. Morphological Facial Index (Total Facial Height / Bizygomatic Breadth)
Divides vertical length (trichion to gnathion) by horizontal cheekbone width (zygion to zygion). Balanced indices measure 1.35 to 1.45, while values below 1.20 denote broad structures and values above 1.55 indicate elongated vertical geometry.
2. Jaw-to-Cheekbone Ratio (Bigonial Width / Bizygomatic Breadth)
Measures bigonial jaw width against cheekbone span to evaluate whether the lower third matches, widens, or tapers relative to the midface.
3. Forehead-to-Cheekbone Ratio (Minimal Frontal Breadth / Bizygomatic Breadth)
Contrasts temple span against cheekbone width, identifying whether the forehead forms an inverted triangular apex or parallel boundary.
4. Mandibular Gonial Divergence & Chin Curvature
Evaluates jawline angles from gonion to gnathion, distinguishing sharp right-angled mandibles (90 to 110 degrees) from soft curved borders (>130 degrees).
Calculations run locally via WebAssembly and WebGL shaders. No biometric photos ever leave the browser sandbox.
7 Classic Face Shapes & Their Geometric Proportions
Anthropometric science classifies human facial variations into seven primary morphological archetypes, each displaying a distinct numeric signature across key dimensional ratios:
1. Oval Face Shape
RATIO: 1.40 - 1.55Features balanced proportions where facial length is roughly 1.5 times cheekbone width. The forehead slightly exceeds the jawline width, converging along a soft, curved jaw toward a gently rounded chin with no sharp angles.
2. Round Face Shape
RATIO: 1.00 - 1.15Displays dimensional parity where facial length closely equals cheekbone width. The jawline forms an obtuse gonial angle exceeding 130 degrees, yielding soft contours with full cheeks and a rounded chin.
3. Square Face Shape
RATIO: 1.05 - 1.20Defined by structural equivalence across horizontal spans: minimal frontal breadth, cheekbones, and bigonial width remain within 5 percent of each other, paired with sharp 90 to 110 degree jaw angles and a flat chin base.
4. Heart Face Shape
RATIO: 1.35 - 1.50Displays an inverted triangular geometry where forehead width substantially exceeds the jawline (ratio above 1.35). The profile tapers dramatically past high cheekbones toward an acute, pointed chin apex.
5. Diamond Face Shape
RATIO: 1.35 - 1.50Cheekbones represent the sole dominant horizontal landmark, noticeably exceeding both forehead width and jaw width. Both upper and lower thirds taper symmetrically toward a narrow hairline and pointed chin.
6. Oblong Face Shape
RATIO: > 1.55Characterized by marked vertical elongation where facial length exceeds 1.55 times cheekbone width. The lateral contours from temples through jaw angles form nearly parallel vertical planes.
7. Triangle Face Shape
RATIO: 1.10 - 1.30Features a broad lower third where bigonial jaw width exceeds both cheekbone width and forehead width. The mandibular base represents the widest facial plane, narrowing upward toward the temples.
Geometric Landmark Analysis vs. Subjective Quizzes
Manual online quizzes and mirror evaluations regularly produce contradictory results due to optical and perceptual distortions:
- Perspective Foreshortening:Smartphone front cameras (24mm to 28mm focal length) introduce barrel distortion at selfie distances, artificially enlarging the central midface and causing cheekbones and jaw angles to recede.
- Lighting and Shadow Variances:Directional lighting creates artificial cast shadows along the jawline and cheek hollows, leading to incorrect visual impressions of bone prominence.
- Self-Perception Bias:Subjective questions regarding jaw "sharpness" lack objective reference frames and vary widely between individuals.
468-point spatial analysis applies affine-invariant 3D normalization, isolating bone structure coordinates independently of skin pigmentation, ambient lighting, or focal warping.
| Evaluation Metric | 468-Point Geometric AI | Subjective Online Quiz |
|---|---|---|
| Measurement Accuracy | Sub-millimeter 3D spatial vectors | Rough visual estimation |
| Distortion Correction | 3D mesh depth normalization | None (uncompensated 24mm warp) |
| Lighting Tolerance | Immune to cast shadows and highlights | Heavily biased by shadows |
| Speed | Under 3 seconds via WebAssembly | 3 to 5 minutes of questionnaires |
| Biometric Privacy | 100% on-device memory, zero uploads | Frequent remote photo storage |
Step-by-Step Guide: How to Detect Your Face Shape in 3 Seconds
Achieving clinical classification accuracy requires following four straightforward photogrammetric steps:
Even Lighting & Direct Frontal Alignment
Position your face directly in front of the camera under diffuse frontal light. Keep hair cleared away from the hairline, temples, and jawline with a neutral expression.
Instant 468-Point Mesh Acquisition
Activate the live camera or upload a photo. The on-device neural network registers facial orientation, projecting a 3D coordinate mesh across your features within milliseconds.
Algorithmic Ratio Computation
The engine derives exact Farkas parameters: facial index, cheek-to-jaw ratio, frontal span, and gonial curvature, mapping your metrics against the seven classic models.
Tailored Styling & Architectural Guidance
Review your biometric breakdown with personalized recommendations for hairstyles, eyewear geometry, and grooming lines engineered to balance your bone structure.
Ready to Discover Your Face Shape?
Scan your face in under 3 seconds using our privacy-first 468-point computer vision engine. Free, private, and client-side with zero server photo uploads.