Deep Learning based Identification of Deepfake Multimedia

Recent advancements in deep neural networks have progressed innovative approaches for creating digital content and it is getting very difficult to ascertain between which content is real and what is deepfake. Attackers are using these developed technologies for tampering with videos and images and disclosing them into social media. These actions are impacting not only individuals in terms of their reputation, mental health, income, etc. but they are also affecting organizations. This research paper investigates deepfake detection in multimedia clips using Deep Learning. Frames are extracted from multimedia clips. Facial detection techniques like region of interest (ROI) and cascading tools like harasses are applied on the extracted frames. A CNN model is trained using processed video frames and used to classify multimedia content into real or fake. Hyper-parameters like batch normalisation, max pooling, and Sigmoid functions are used for fine-tuning of the model to achieve better accuracy. Results show CNN provides good accuracy for deepfake videos detection.