Detection of Facial Deep Fake Images using CNN Architecture

Facial deepfake images are manipulated images where a person’s face is altered with computer generated faces, or some changes are made to the facial features of the images that appears realistically. In the developing world, many technologies are used for content creation and manipulation. Those deepfake images are developed using deep learning techniques such as the Generative Adversarial Network (GAN). Those synthesized (fake) images with inappropriate content can be used on social media platforms, which leads to severe problems and causes harm to the individual’s reputation. An efficient and effective facial deep fake detector is necessary to solve the problems. Various deep-learning techniques are used to detect facial deepfake images. This paper proposes a model based on the convolutional neural network (CNN) architecture that proficiently detects facial deepfake images. The proposed model based on the convolutional neural network includes convolutional layers, DenseNet121, pooling layers, and an Adam optimizer. where convolutional layers are very crucial for capturing spatial features, DenseNet121 is able to implement a feed-forward information flow paradigm, allowing features to directly propagate from earlier layers to later layers, pooling layers can down-sample the spatial dimensions and retain the spatial features. The Adam (Adaptive Moment Estimation) optimizer is an optimization algorithm that can make the model work efficiently and helps to accelerate the optimization process. This model can improve the detection of fake facial images more than any other existing model.