Deepfake image detection using a deep hybrid convolutional neural network

Abstract

Deepfake refers to a category of fake and artificial data in which fake content is produced based on existing content. This content can include image, video and audio signals. Deepfake production is based on deep generative networks that manipulate data or produce fake images and videos. In recent years, many studies have been conducted to understand how deepfakes work, and many methods based on deep learning have been introduced to identify videos or images produced by deepfakes and distinguish them from real images. In order to improve the accuracy of deep-fake detection and simultaneously use the capabilities of different types of convolutional neural networks, in this article, a hybrid model is presented using four convolutional neural networks: DenseNet201, EfficientNetB2, Inception-ResNet-V2, and ResNet152. turns Relying on the high capabilities of these networks in extracting effective features from the input image, the proposed model is able to simultaneously recognize whether the input image is deep or not by these four models. The results presented on the three databases of 140k real and fake faces, DFDC faces and Deepfake and real images indicate the improvement of the results compared to the existing models.

Keywords

• deepfake
• deep learning
• deep convolutional neural network
• accuracy
• fake images