A Survey on Deep Fakes Detection

A Survey on Deep Fakes Detection

Author Name : Sai Siddhu Gedela, Keerthika Challa, Hymavathi Kusumanchi, Pavan Gurrala, Suvarna Daki, Ms. Y. Nagamani

ABSTRACT

The modified photos and movies known as "deepfakes" are produced by executing a face swap using a variety of AI technologies. This gives the appearance that someone stated something they did not say or that they are someone they are not. There are significant consequences from employing the deepfake technology incorrectly. These deep fakes have harmful effects, such as inflaming political tension, simulating terrorist attacks, and undermining people's dignity and reputation. With the naked eye, it is challenging to see the effects of this deep-fake technology. Therefore, a deep learning- based approach that can effectively distinguish deepfake videos is created in this study. The suggested method divides the input video into frames, and then utilizes a Res-Next pretrained Convolutional neural network to extract features from those frames. The retrieved features are then used to train a Long Short-Term Memory (LSTM) model to determine whether or not the user-supplied input video has been altered in any way. The model is trained with a balanced dataset to improve performance on real-time data. The Deepfake detection challenge dataset was used to train the model. The outcomes of the suggested model can be predicted more accurately. This approach can thereby counteract the hazards and danger that deepfake technology has brought to society.

Keywords—Deepfakes, Deep learning, Artificial Intelligence (AI), Long Short Term Memory (LSTM), Res-Next Convolutional Neural Network