DeepGuardNet: A Novel CNN Architecture for DeepFake Image Detection

We are in the world where information is the ubiquitous but the authentication of that information is cumbersome. Deepfake technology have achieved a tremendous growth in the digital era. Deepfake is a synthetic media audio, video or images that appear to be realistic though they are fake or fabricated. Deepfake are created by Generative AI techniques that understand the probabilistic distribution of the data. Appropriate detection systems are necessary to prevent the dissemination of misleading information and guarantee the authenticity and integrity of data. Detecting deepfake contents in this digital era is very challenging due to the realistic nature of fake images. In this paper, we present an enhanced CNN architecture, DeepGuardNet, a deepfake detection model that is simple and effective for determining whether images are real or fake. DeepGuardNet is a straightforward, sequential, and robust network designed for deepfake recognition and detection. Additionally, our network has an enhanced ability to detect tampered content with fewer parameters due to the use of separable convolution. In this work, we utilize depthwise separable convolution to efficiently extract deepfake features. The DeepGuardNet architecture effectively captures deepfake image features in both the spatial and depth dimensions. Experimental study on Celeb-DF dataset demonstrated the competence of the proposed method with an accuracy of 91% when compared with conventional methods. The proposed DeepGuardNet architecture is productive in terms of the better feature extraction and reduced computational complexity.