The proliferation of manipulated multimedia content poses a significant threat in an era heavily reliant on social networks as primary information sources. Despite numerous countermeasures targeting specific attack types, the seamless nature of image manipulation challenges the differentiation between authentic and altered visuals. This study aims to detect fake images generated by StyleGAN2-ADA using watermark analysis and image content analysis techniques. The first experiment evaluates the performance of watermarking techniques in the spatial (Least Significant Bit (LSB)) and frequency (Discrete Cosine Transform (DCT)) domains using real-life imagery. Then, watermarked images are used as input to the StyleGAN2-ADA model to generate synthetic counterfeits. The second experiment assesses the effectiveness of content-based analysis techniques in distinguishing between authentic and forged images, including Error Level Analysis (ELA), perceptual hashing, and a pre-trained Convolutional Neural Network (CNN) model. Finally, the third experiment integrates the findings from the two previous experiments to provide a reliable determination of image authenticity. The results show that by leveraging watermark-based and content-based detection, the proposed framework achieves high accuracy in identifying fake images generated by StyleGAN2-ADA.
