Detecting Digital Deception: A CNN-RNN hybrid Approach of Deepfake Detection

Deepfake technology has forced digital deception to create high demand for detection tools. This paper presents an approach for deepfake video detection using Convolutional Neural Network (CNN) and Recurrent Neural Network (RNN) architecture that targets the artifacts present in the output from the generative models including Generative adversarial network (GAN). This approach looks at many strategies in terms of frames, among them for frame comparison, feature extraction based on CNNs, and for the purpose of temporal analysis CNN with long-short-term memory (LSTM) networks. The described model is trained on the set of the genuine vs. forged images and videos and demonstrates quite stable results with respect to digital forging detection. It is discovered that this proposed model achieves higher accuracy than previously used detection approaches in addressing face-swapping and face-reenactment deepfakes. The findings of this research demonstrate that CNN & RNN based deepfake detection is promising for media forensics, as it advances multimedia security and digital media credibility. This proposed method shows an accuracy of 81% in the Deepfake Detection Dataset (DFDS), respectively, with a very reduced number of sample size of ⩽ 100 samples(frames). This promises early detection of fake contents compared to existing modalities.

Перевод пока недоступен