Deepfake technology has significantly evolved with advancements in deep learning and Generative Adversarial Networks (GANs), enabling the generation of highly realistic synthetic facial images and videos. While these developments offer innovative applications in media and entertainment, they also pose serious threats such as misinformation, identity impersonation, and digital fraud, making the detection of manipulated facial content a critical challenge in digital media security. This paper presents a deepfake face detection system based on the Inception-ResNetV2 architecture that uses transfer learning for efficient feature extraction. The proposed model performs binary classification to distinguish between real and fake facial images and incorporates a localization mechanism to visually highlight manipulated regions. The system was implemented using TensorFlow and OpenCV and deployed through a web-based interface to enable real-time detection. Experimental evaluation conducted on a large-scale dataset demonstrated high classification accuracy and reliable performance across training, validation, and testing phases. The results indicate that the proposed framework provides an effective, practical, and scalable solution for deepfake detection and digital media authentication, contributing to improved cybersecurity and trust in digital content.
Maddala et al. (Sun,) studied this question.