GAN (Generative Adversarial Network)
Definition
A neural architecture introduced by Goodfellow et al. (2014) consisting of a generator and discriminator trained in opposition. The generator learns to produce samples indistinguishable from the training distribution. Applied to handwriting and signature generation with conditioning on reference examples from a specific writer.
- Origin
- Ian Goodfellow et al., 2014
- Components
- Generator network and discriminator network trained jointly
- Training goal
- Generator output becomes indistinguishable from real data to the discriminator
- Forensic use
- Conditioned generation of handwriting or signatures from a writer's reference samples
Common questions
How does a GAN differ from other image-generation methods for forgery?+
A GAN is trained specifically against a discriminator that learns to spot fakes, so the generator improves at defeating detection during training itself. This adversarial loop is what makes GAN output harder to flag by eye than output from a single generative model trained without a competing critic.
Can a GAN-generated signature fool a forensic document examiner?+
It can defeat casual visual inspection, but stroke dynamics, pen pressure sequencing, and pixel-level statistical artefacts left by the upsampling layers often still differ from genuine writing, which is why examiners combine visual comparison with instrumental and statistical analysis rather than eye alone.
Why does a GAN need a discriminator instead of just training the generator alone?+
Without a discriminator providing an adversarial signal, the generator has no way to learn what makes output realistic beyond matching simple statistics; the discriminator supplies a moving target that forces increasingly convincing output.
Related terms
- Adversarial Attack
- In AI detection contexts, a forgery specifically constructed to defeat a trained detector by incorporating small perturbations that cause the detector's classifier...
- Automatic Signature Verification (ASV)
- Algorithmic systems that compare a questioned signature against a reference set using extracted biometric features (stroke curvature, velocity profile, pressure distribution). Designed...
- C2PA (Coalition for Content Provenance and Authenticity)
- An open technical standard that embeds cryptographically signed provenance assertions into media files at the point of capture or editing. A C2PA...
- Content-Aware Fill
- An image editing feature (Photoshop from CS5; AI-driven from version 22.0) that replaces a masked region with synthesised pixel content matching the...
- Diffusion Model
- A generative neural network architecture (Ho et al., 2020; Stable Diffusion, Rombach et al., 2022) that learns to reverse a noise-addition process...
- ESDA (Electrostatic Detection Apparatus)
- A non-destructive technique for detecting indented impressions on paper caused by writing pressure. A physical ink signature on paper produces detectable ESDA...
- GAN Spectrum Fingerprint
- Periodic artefacts in the Fourier frequency spectrum of images generated by GAN architectures with upsampling layers. First reported by Zhang et al....
- JUMBF (JPEG Universal Metadata Box Format)
- A container format (ISO/IEC 19566-5) used by the C2PA specification to embed provenance manifest data within JPEG image files. The JUMBF box...
- PRNU (Photo Response Non-Uniformity)
- The unique pattern of pixel-level sensitivity variations in a camera sensor, used as a device fingerprint. A genuine camera-original image carries the...