Boundary-Based Fake Face Anomaly Detection in Videos Using Recurrent Neural Networks

With the surge in the videos over the internet and the spread of digital communications, the attackers are greatly benefited by creating and publishing the fake distorted videos and images known as Deepfakes. The recent trends, Machine Learning powered applications are highly implied to create deepfakes that leads to fallacious arguments. To detect such an instance, there are several existing techniques that uses convolutional neural networks and machine learning methods which are limited to identify the larger details of the deepfake created. We propose a novel anomaly detection technique for face areas in videos using recurrent neural networks based on boundary marking and hashing technique which is efficient in detecting the deepfakes over a smaller face boundary regions. The proposed method is evaluated, and the results show promising performance for deepfake detection in the face area based on the boundaries. Our proposed method revealed good accuracy in detecting the smaller frame 8  $$\times $$  8 raster based anomalies in videos.