.png)
Automatic Focus Detection
Advanced 3-metric ensemble algorithm for evaluating facial image sharpness and blur.
Automatic Focus Detection evaluates the sharpness of an uploaded face image using a multi-metric ensemble approach instead of a single gradient measure. This ensures high accuracy and consistency across different lighting conditions, distances, and blur types (defocus vs. motion blur).
Algorithm Overview
The system uses a 3-metric ensemble, resulting in a final score between 0–100 (where higher means sharper).
| Metric | Weight | What it detects |
|---|---|---|
| Tenengrad (Sobel gradient energy) | 60% | General sharpness, resolution-independent |
| Crête blur score | 30% | Content-independent perceptual blur |
| FFT high-frequency ratio | 10% | Motion/directional blur invisible to gradient methods |
Classification
The final focus_quality_score (0–100) is translated into five human-readable categories:
- ≥ 75:
sharp - ≥ 55:
acceptable - ≥ 35:
slightly_blurry - ≥ 15:
blurry - < 15:
very_blurry
Facial Context Application
During detection, the Tenengrad gradient evaluation specifically weights pixel values nearest to the eye landmarks 1.5× higher than other areas. This ensures the biometric pipeline receives photos where the eyes—which are crucial for precise InsightFace embedding extraction—are heavily in focus.