An Effective Technique for Deepfake Video Detection

As fake videos cause numerous problems affecting people's lives in various fields, they have received increasing attention. In this paper, we present a successful method for detecting fake videos based on the Scattered Wavelet Transform (SWT) and a pre-trained deep learning model, EfficientNet-B0. Several experiments were conducted on the fake video detection dataset to evaluate the effectiveness of the proposed method. The Deepfake Detection database (DFD) database was used. The model was tested by adding noise to images to evaluate its robustness and accuracy under various noise conditions. It was tested on salt-and-pepper noise and white Gaussian noise, and horizontal misalignment noise, a type of noise commonly used in fake video detection, was applied. 98% accuracy was achieved using the noise.