Detecting AI-generated images with CNN and Interpretation using Explainable AI
Bharathi Mohan G, R Prasanna Kumar, Akilesh rao S, Mandava Sukesh, D M Abinandhini, Y Jaikanth
Abstract
The rapid proliferation of digitally manipulated images in various media outlets has necessitated the development of robust and reliable methods for distinguishing between authentic and synthetic visual content. In response to this imperative, this research leverages state-of-the-art deep learning techniques and an extensive dataset of ‘real’ and ‘fake’ images to create an AI model for image identification. The study encompasses a meticulous data collection process, preprocessing steps, the selection of the DenseNet121 model architecture, model training, and evaluation of the model's performance through a range of metrics, including accuracy (93.42), precision, recall, and the Area Under the ROC Curve (AUC). To enhance the transparency and interpretability of the model, employed the Gradient-weighted Class Activation Mapping (Grad-CAM) technique, enabling to visualize the regions within images that influenced the model's decision-making. The research provides valuable insights into the development and evaluation of AI systems for image identification, particularly in the context of distinguishing between real and fake content. Furthermore, the interpretability techniques shed light on the model's decision-making process. This research contributes to the field of image authentication and paves the way for enhanced image classification in the era of synthetic media.