Novel 59-layer dense inception network for robust deepfake identification

The exponential growth of Artificial Intelligence (AI) has led to the emergence of cutting edge methods and a plethora of new tools for media editing. The use of these tools has also facilitated the spread of false information, propaganda, and harassment through the creation of sophisticated fake video and audio content, commonly referred to as deepfakes. While existing efforts exist for identifying deepfakes videos, they have received less attention when it comes to social media videos. This paper presents 59-Layer Fake Dense Inception Network (FDINet59), designed to detect deepfakes contents. The dataset generated by Multi Task Cascaded Convolutional Networks (MTCNN) crop for training the system and evaluated its ability to spot deepfakes on datasets. The results show that FDINet59 provides impressive performance in identifying deepfakes material, achieving a maximum accuracy of 70.02% with a loss of 0.688 log units while using the training dataset. The ability of FDINet59 to detect deepfakes content generated by auto encoders and Generative Adversarial Network (GAN), commonly used to create deepfake videos. The results show that FDINet59 is 94.95% accurate with a log loss of 0.205. The proposed model can play an important role in preventing the spread of deceptive deepfake videos on social media. The development of more sophisticated deepfake detection algorithms is crucial to counter the negative impacts of this technology on society.