Robust Fake Image Detection Using a Regularized Convolutional Neural Network

Abstract

Detecting fake images has become a critical challenge in the era of advanced image generation and manipulation techniques, such as those produced by Generative Adversarial Networks (GANs). In this study, we propose an advanced Convolutional Neural Network (CNN) architecture integrated with regularization techniques to improve the accuracy, generalization, and robustness of fake image detection. Our model is evaluated against five state-of-the-art approaches, including methods based on visual artifacts, forensic classification, and spectral inconsistencies. The proposed model achieves a validation accuracy of 91%, significantly outperforming existing techniques, with the lowest validation loss of 0.27 and superior precision, recall, and F1-score metrics. Through comparative analysis, we demonstrate the model’s ability to effectively address challenges in detecting subtle manipulation artifacts across diverse datasets. This work highlights the potential of our method as a reliable solution for ensuring the integrity of visual media and combating the proliferation of fake images. Future directions include further optimization for real-time applications and adaptation to evolving manipulation techniques.