Deepfake Image Detection using a Unified Inception-V3-Xception Net Framework

Advancements in multimedia content generation and modification have reached a level of realism where distinguishing between authentic and synthesized visuals is challenging. A leading technology in this field is deepfake, a generative AI model capable of altering facial features with remarkable realism. This capability has seen extensive application in fields like television broadcasting, video games, and cinema, especially in enhancing visual effects. Unfortunately, deepfake technology is also misused to create deceptive content by replicating the likeness of well-known individuals, often for spreading misinformation. Consequently, there is a growing focus on deepfake detection, leveraging deep neural networks (DNNs) to identify and categorize deepfakes. In this study, a new classification framework is proposed that combines the strengths of two deep learning models, Inception-V3 and Xception Net, to distinguish genuine images from those generated by deepfake technology. The model is trained and tested using a dataset from Kaggle, and experimental results highlight the significant effectiveness of this approach in comparison to existing techniques.