AI-Driven Detection of Deepfake Profile Pictures on Social Media

Abstract

It is becoming easier to create highly realistic but entirely fake human faces thanks to the growing use of Generative Adversarial Networks (GANs). Because of this, the number of deepfake profile pictures on social media has gone up a lot. These images are so convincing that it's hard to tell they aren't real. Now, they are being used more often to create fake accounts that spread false information, commit fraud, or manipulate people's beliefs. This paper looks at how GAN-generated photos are used in fake social media profiles and introduces an AI method for detecting them.
We begin by examining the typical features of deepfake profile pictures, such as eyes that don't align properly, mismatched backgrounds, and odd metadata problems. We then compare these with real user profile images. By using a large collection of labeled images, including both fake ones created by GANs and genuine ones, we train various deep learning models like CNNs and Transformer-based models to identify subtle visual and pixel clues that GANs often leave behind. We also apply image forensic techniques and look at social connections to enhance the accuracy of deepfake detection.Our findings show that we can accurately detect fake profile pictures, especially when using both image analysis by AI and the behavior of the accounts linked to those pictures. The study also points out the big challenge in the ongoing battle between AI that creates fake content and the tools that try to stop it. It stresses the importance of having automatic security measures to help social media stay safe from false identities and organized efforts to spread misinformation.