Repository navigation
v2.8.0
Adds AI-generated image detection, flagging statistical signatures consistent with GAN/diffusion-model generation rather than a real camera capture — a complement to the existing tamper-focused detectors.
Added
- Frequency-domain spectral analysis detecting periodic artifacts characteristic of GAN upsampling layers
- Global PRNU-absence check: flags images lacking any coherent camera sensor noise fingerprint across their full extent
- Combined "likely AI-generated" score feeding into the existing suspicion scoring pipeline
- New config options: spectral-analysis sensitivity, PRNU-absence threshold, enable/disable switch
Notes
- Based on classical statistical signal analysis, not a trained classifier — results are signals consistent with AI generation, not a certain verdict, and accuracy may degrade as generative models evolve
- Cross-platform (macOS + Linux), no breaking changes to existing CLI usage
Full Changelog: (Generate release notes butonuna basarsan GitHub otomatik commit karşılaştırma linkini ekler)