An Innovative Model of Deepfake Detection in Video Using 3D EfficientnetB7 with Spatial Attention
T. M. Rajesh, S. Maruthuperumal · 2025
Nowadays, the face-swapping Deepfake models are relatively spread, producing a large amount of realistic fake videos that create concerns for countries and people's privacy. Because of the fake video's negative impacts on the world, differentiating the Deepfake and the original videos has become a significant problem. The Deepfakes can be leveraged for slander, political advantages, and to defame the public figure's reputation. Despite Deepfake's imperfections, people find it hard to differentiate among manipulated and authentic videos and images. As a result, it is significant to have automated models that effectively and accurately categorize the digital content's variability. Deepfakes enable the automatic creation and generation of (false) video content, e.g., via the Generative Adversarial Networks (GANs). The technology of Deepfake is a controversial mechanism with large problems affecting society. Distinct current Deepfake identification strategies employ the video's single frames and concentrate on the image's spatial data to infer the video's authenticity. Several effective techniques utilize the manipulated video's temporal inconsistencies. Nevertheless, the experiment highly concentrates on the spatial features. To eliminate these problems, this work develops a new mechanism for detecting Deepfakes. The novelty of the developed work is to identify the Deepfakes from the video thus helping to prevent fake news propagation and also improving the media's credibility. In the beginning, the necessary videos for implementing the model are obtained from the available resources. Further, these images are passed to the 3D EfficientnetB7 with Spatial Attention (3DENetB7-SA) model for detecting the input videos as real or fake. Finally, the necessary validations for the implemented model are carried out with the support of important performance measures over existing models to ensure the efficacy of the suggested Deepfake detection framework.