An Automated DeepFake Video Detection System Utilizing Depthwise Separable Convolutional Neural Networks
Jalaj Patwal, Akash Chauhan, Aman Rawat, Shruti Dabral, Tamanna Tamanna · 2024
A DeepFake is a manipulated form of media created using "Deep" learning. Its ability to spread misinformation and threaten personal rights presents a threat. Additional outcomes of DeepFakes include blackmailing, financial scams and harming the reputation of an individual. All these consequences lay emphasis on the significance of its detection. The research aims to detect deepfakes at a more accurate level. The study revolves around Convolutional Neural Network (CNN) architecture. We examined the accuracy of the architecture. The architecture was set up as our base. However high levels of accuracy were not achieved. To enhance accuracy, the research has implemented Xception model architecture and trained it on a huge dataset. Thus we were able to make enhancements in the accuracy. To further achieve a greater precision an optimizer was used. Specifically, "Adam" optimizer was involved in the fine-tuning process at different learning rates. Thus, the highest level of accuracy 99% was achieved.