Deepfake Detection Using CNN Trained on Eye Region

In this work, we will develop a simple convolutional neural network to detect deepfakes in videos on a frame-by-frame level, focusing on the region around the eyes. Since deepfakes are increasingly being created using forms of CNN, it should be possible to also detect deepfakes using CNN. OpenCV allows for frame extraction from videos, while also allowing image cropping. The well-developed Multitask Cascade Neural Network (MTCNN) is a stacked neural network for face detection and alignment. MTCNN is used for high accuracy face detection to greatly reduce false positive images in the dataset. Finally, a region around both eyes are cropped, with extra padding, to be used as input to train a CNN, using returned coordinates from MTCNN for the eyes. This research will focus on measuring if the eye region can be a useful area of interest for comparing original videos to deepfake videos.