How I Built AI Face Shape Detection Using Facial Landmarks
I’ve been experimenting with facial landmark detection and proportional analysis over the past few months, and built a small project that classifies face shapes from a single image.
The system maps facial geometry using landmark points across the jawline, cheekbones, forehead, chin, and temples, then calculates proportional ratios to classify structures like oval, round, square, oblong, diamond, and triangle.
One thing I found interesting is how often people misidentify their own face shape manually — especially oval vs oblong and round vs oval.
I also started exploring:
facial symmetry analysis
hairstyle compatibility
beard balance
glasses recommendations
proportional face thirds
The biggest challenge was handling:
head tilt
inconsistent lighting
partial hair occlusion
beard interference with jaw detection
Would love feedback from people working with computer vision, Mediapipe, or facial analysis systems.
Project:
faceshapedetector.store
