Detects whether an image is a real photo or a photo of a screen. No deep learning — 76 FFT/moiré features (pixel-grid aliasing, refresh banding, glare) into gradient-boosted trees, trained on real recaptures. 94.7% grouped 5-fold CV, ~0.5s per 12MP image on CPU. Flask web demo included.
python opencv flask computer-vision scikit-learn fft image-forensics fraud-detection gradient-boosting moire-pattern screen-detection recapture-detection
-
Updated
Jul 2, 2026 - Python