Detection of Artifacts in AI-Generated Portraits Using Frequency and Texture Analysis (FFT, DCT, LBP/GLCM)

In this article, we present interim results from ongoing research aimed at identifying differences between real and AI-generated portraits through analysis in the frequency and texture domains. Three methods are examined: Fast Fourier Transform (FFT), Discrete Cosine Transform (DCT), and statistical texture descriptors based on Local Binary Patterns (LBP) and the Grey-Level Co-occurrence Matrix (GLCM). Using a controlled set of image data – a real portrait, its AI-generated clone, and a retouched version – we demonstrate the processing workflow, visualization, and interpretation of results. The aim of this work is to verify whether visually subtle differences between real and synthetic images correspond to measurable structural differences in alternative image representation domains. The article presents a methodological framework and example results of a pilot study; more extensive experiments on a larger dataset and the inclusion of additional analytical tools are planned in subsequent phases of the ongoing research.