Deepfake Detection in Picture

With the growing prevalence of hyper-realistic counterfeit images, particularly Deepfake and Face2Face, accurately identifying manipulated facial expressions has become a critical challenge in digital forensics. Traditional forensic methods often fail due to the substantial degradation of image data during manipulation processes, necessitating novel approaches. This paper introduces two lightweight deep learning networks tailored to effectively detect facial expression anomalies while focusing on mesoscopic image structures. Unlike conventional methods that rely on complex architectures, our proposed networks employ a minimal number of layers to prioritize computational efficiency and structural precision. We evaluate these networks on both established datasets and a newly curated online photo dataset to assess their performance comprehensively. The suggested strategy yields an impressive detection rate of 97% for Deepfake, according to the experimental results and 94% for Face2Face, underscoring its robustness and efficacy. This work contributes a streamlined yet highly effective approach for detecting manipulated facial images, offering promising advancements in combating visual misinformation in digital media.