Advanced Deepfake Detection Using Inception-ResNet-v2

In recent years, the prevalence of manipulated videos, commonly referred to as “deepfakes”, has surged, impacting various industries worldwide. These videos are generated by software based on deep learning models. In the near future, making deepfakes is likely to become simpler than it is now, as the current process demands a hefty amount of data and expensive computing power. The repercussions of this technology’s misuse, including the spread of false information and potential threats to national security, pose significant concerns such as libel, sabotage, and blackmail. This study proposes a system employing MTCNN for facial extraction in videos, feeding the data into a sequential model featuring Inception-ResNet-v2 and additional layers. Rigorous testing with diverse video samples demonstrated favorable results, highlighting the model’s effectiveness and accuracy. The proposed model archived better accuracy 91.41% as compared with existing CNN and Xception net accuracy. This abstract encapsulates the key aspects of the research, from the problem statement to the proposed solution and successful outcomes, contributing valuable insights into addressing the challenges posed by deepfake technology.