Face manipulation¶
Face manipulation encompasses techniques for synthesizing, editing, or altering facial images and videos using generative models, face-swapping algorithms, and digital forensics. This includes deepfake synthesis, facial reenactment, attribute editing, and expression transfer.
Techniques¶
- Face-swap: Identity transfer—replacing one person's face with another in an existing video while preserving lip-sync and expressions
- Facial reenactment: Transferring facial expressions and head movements from a source video to a target face
- Attribute editing: Modifying facial attributes (age, expression, gender) while preserving identity
- Face synthesis: Generating completely synthetic facial images or videos using GANs
- Expression transfer: Copying emotional expressions from source to target
Detection approaches¶
Face manipulation detection relies on identifying artifacts introduced by the manipulation process: frequency-domain anomalies, temporal inconsistencies, biological signal disruption (eye blinking patterns, heartbeat), and spatial-temporal feature discontinuities. Deepfake Detection: A Systematic Literature Review systematically reviews 112 detection methods showing deep learning (especially CNNs) dominates the field.
Key papers¶
- Recurrent Convolutional Strategies for Face Manipulation Detection in Videos — Temporal analysis for manipulation detection in video; recurrent CNNs combined with face alignment detect Deepfake, Face2Face, and FaceSwap with 94–97% accuracy.
- FakeCatcher: Detection of Synthetic Portrait Videos using Biological Signals — Detection via biological signals (photoplethysmography); exploits spatial coherence and temporal consistency of blood flow patterns
- DeepFakes and Beyond: A Survey of Face Manipulation and Fake Detection — comprehensive survey of four manipulation types (entire face synthesis, identity swap, attribute manipulation, expression swap), techniques, and detection methods
- Deepfake Detection: A Systematic Literature Review
- FaceForensics++
- DFDC