1. 1. Face detection and alignment
The model finds faces in the frame and aligns them using eye and mouth landmarks, reducing angle and scale differences.
2. 2. Feature vector and score
The aligned face becomes a numeric vector. Vector distance is converted into a similarity score; a high score is a strong indicator, not identity verification.
3. 3. Limits
Accuracy drops with blurry, very small or heavily occluded faces. Use a sharp, front-facing frame for the best result.
#Face search#Technology