Tracking Realness Using Smart Technology

Abstract

Computer vision technologies along with deep learning algorithms have boosted the creation of deepfakes by making manipulated multimedia data more authentic. Extremely believable forged media objects now enable scammers to distribute deceptive information while executing financial schemes. This paper presents Tracking Realness Using Smart Technology as a complete image and video authentication system which operates at scale. Our system follows successive steps beginning with video-to-image conversion through discrete frame extraction followed by MTCNN and Azure Vision API face detection for dominant facial identification followed by dataset development for EfficientNetB0 CNN binary classification training. The deployed model operates through a web application constructed with Django programming together with Docker container distribution features for time-sensitive operations. Results of Deepfake Detection Challenge, FaceForensics++ and Celeb-DF datasets validate high accuracy of detection of the system with low power consumption. A solid foundation for enhancing automated approaches to deep fake detection and multimedia security systems.