Detection of fake images generated by GAN using transfer learningin convolutional neural networks

Generative Adverserial Network (GAN), which is one of the popular networks of deep learning methods, is known as a productive deep learning model. It consists of two parts, the generator and the discriminator. There are very successful examples of GAN production data, known as synthetic or artificial data, in the literature. It is not possible to detect synthetic face data, which are known to be used in quite different fields, by human perception. Within the scope of this thesis, a model including a Laplace filter and a new dissimilarity-based convolution layer is proposed for a network that distinguishes synthetic and real face images, in which Convolutional Neural Network (CNN) models, which are widely used in the literature, are used as feature extractors. In this method, Densenet121 model is used as feature extractor and features are given to machine learning models in order to classify them. The new dataset created within the scope of the thesis study was tested on different machine learning models, and the highest success was obtained from LR and KNN models at 99%. The obtained results show that the proposed method can be used to detect fake faces.