The rise of generative AI presents a profound duality. On one hand, it offers a powerful solution to data scarcity and privacy challenges in biometrics. On the other, it is weaponized to create deepfakes that threaten digital integrity. Existing detectors for these deepfakes are brittle, failing against real-world transformations and novel generative models. This dissertation confronts this duality head-on. First, I establish the viability of synthetic data for building fair and private biometric systems. Second, to counter the malicious use of this technology, this dissertation develops deepfake detectors designed to be robust, generalizable, and efficient by construction. My work introduces novel, lightweight feature sets on different cues (e.g., colour cue-based Relative Chrominance Difference, Gradient features, Depth cues, etc.) that are inherently resilient to OSN transformations and improve generalisation to unseen forgeries. Whereas, accomplished results confirm state-of-the-art performance, achieving high accuracy in challenging real-world scenarios with a significant reduction in model complexity, my current and future work focuses on achieving superior generalisation while being OSN manipulation resistant.

The rise of generative AI presents a profound duality, or a ''Janus Face,'' for digital society. On one hand, its ability to synthesize hyper-realistic faces offers a powerful solution to long-standing privacy and data scarcity challenges in biometric systems---a promising but underexplored application. On the other hand, this same technology is weaponized to create 'deepfakes' that fuel misinformation campaigns on Online Social Networks (OSNs), posing a significant threat to digital integrity.

The rapid advancement of Generative AI, especially Generative Adversarial Networks (GANs), has increased the issue of fake news on Online Social Networks (OSNs) by generating deceptive face images for social media profiles. Although existing detection methods are accurate, their effectiveness decreases when images are post-processed, which is common on OSNs. In this paper, we present LPQ-Net, a model combining Local Phase Quantization (LPQ) for feature extraction with a CNN-based classifier. We explore two variants: one sets a new benchmark in detecting StyleGAN2-generated images, and the other excels in identifying images shared on Facebook, WhatsApp, and Instagram. LPQ-Net also operates with minimal parameters, outperforming state-of-the-art methods and making it ideal for resource-constraint applications. Furthermore, our solution demonstrates its effectiveness by performing exceptionally well in detecting images generated by various Diffusion models. We further show that incorporating LPQ features into fine-tuned classifiers like ResNet50, ResNet101, InceptionV3, and DenseNet121 significantly improves performance.

Creating hyper-realistic synthetic images has become effortless with tremendous development in Generative Artificial Intelligence technologies. Generative Adversarial Networks (GAN) generated synthetic images, especially non-existent face images that are visually indistinguishable from real faces, pose a severe social threat by enabling misinformation dissemination, often over online social networks and through fake social profiles. In spite of successful solutions being reported in the recent literature for detecting GAN-generated synthetic images, the performance of such schemes degrades considerably with the launch of post-processing attacks. In this work, we employ gradient of an image as the key component to detect synthetic images. According to our results, gradient proves to be a considerably efficient image derivative for synthetic image detection as well as to achieve robustness against post-processing attacks. We explore two different gradient operators and design four unique deep learning-based detection networks utilizing different gradient-based feature sets. Our solution achieves state-of-the-art (SOTA) detection accuracy (above 99%) on the test set consisting of STYLEGAN2 images and outperforms SOTA solutions for detecting post-processed and compressed images.