A Hybrid approach for Deepfake Detection using CNN-RNN
Sonali Mallinath Antad, Vaishnavi Arthamwar, Rohan Kishor Deshmukh, A. Chame, Harshita Pawan Chhangani · 2024
Reliable deepfake detection of falsified videos generated by deep learning is still a pressing challenge with its growing popularity. Deep adversarial neural networks, which educate on film and target faces to ahead facial resources and facial expressions to targets, underneath DF can without difficulty idiot facial recognition structures the usage of revealed, 3-D photos mask, or video recordings from the valid consumer’s face to sensors. We make significant contributions, including building adaptive datasets, comparing primitive and deep models for training, and tunable CNN-RNN tools. The overall evaluation model shows that the results are better than the findings of the most advanced methods. It was acknowledged that Deep faking today greatly impacts our environment; It is done by placing faces on original images/videos using deep neural networks. Together with the sharing of other misinformation via digital social networks, deepfake has formed digital fakery, which has become a real problem of negative social impact and as such there is a great need for successful measures to be taken to examine Deep Fakes [1]. The approach in this research generalizes by capturing physical cues in body shape and size. Our method combines CNN for spatial analysis, capturing facial features and skin texture, and LSTM and RNN for temporal analysis, monitoring expression progress. This method provides stable performance on complex deepfake videos. The modular design allows for future expansion, such as the integration of audio analysis.