AI-generated image detection using Machine Learning techniques
AI-generated image detection using Machine Learning techniques
J. Chorazy,Łukasz Dąbała
TLDR
This work aims to implement or adapt three contemporary methods for generated image detection: aPhoto-Response Non-uniformity (PRNU) extractor paired with acustom CNN, an Error-Level Analysis (ELA) extractor coupled with the same custom CNN, and an adaptation of the LatentReconstruction Error framework LaRE2.
Abstract
Rapid developments in the field of generativeartificial intelligence have enabled the creation of photorealisticimages that are becoming increasingly difficult to distinguishfrom real photographs. This work aims to implement or adaptthree contemporary methods for generated image detection: aPhoto-Response Non-uniformity (PRNU) extractor paired with acustom CNN, an Error-Level Analysis (ELA) extractor coupledwith the same custom CNN, and an adaptation of the LatentReconstruction Error framework LaRE2. Each of the methodshas been thoroughly tested on a custom dataset, which wasconstructed by combining part of the Tiny GenImage datasetwith images generated by a state-of-the-art transformer-diffusionmodel named FLUX. The data, balanced evenly between realand fake images, has been separated into five subsets, eachrespective to one of the included models: ADM, BigGAN, FLUX,Midjourney, and Stable Diffusion v1.5. Experiments were conductedin three different categories, to ensure proper validationof performance of the tested methods.