Fake Face Detection with Separable Convolutions

Fake images bring fake news that causes harmful consequences for victims, many dangerous effects on society, and causes economic damage. For instance, a fake face embedded into an image can be perilous because it can deceive and mislead people, leading to false identification, impersonation, and even fraudulent activities. However, with great support from information technology, such fake images can be easily created by embedding and retouching the victim’s face so sophisticated that it is difficult to distinguish with the naked eye. This study proposes approaches to detect fake faces in images using well-known convolutional neural networks, including Separable Convolution-based architectures, Inception, EfficientNet, MobileNet, and Xception, to perform fake face detection in images. The study is carried out on datasets of about 20,000 images with deep learning techniques. Experimental results reveal that separable convolution-based architectures have performed best and better than some previous studies.