VISTAS: REAL-TIME DEEPFAKE DETECTION USING MOBILENETV2 FOR AUTOMATED FACIAL FORENSICS

The rapid rise of hyper-realistic synthetic media, commonly known as Deepfakes, has
posed significant challenges to the integrity of digital content. Traditional verification methods relying
on human visual inspection are increasingly inadequate, as Generative Adversarial Networks (GANs)
can produce forgeries that are nearly indistinguishable from authentic images. Manual detection is
time-consuming, error-prone, and insufficient to curb misinformation, identity theft, and cyber fraud.
To address these issues, an automated and intelligent forensic system is required to ensure accurate

and efficient media authentication. This paper presents VISTAS (Visual Integrity Security & Track-
ing AI System), a real-time deepfake detection framework that leverages deep learning techniques

for facial forensic analysis. The system captures image inputs via a secure web interface, extracts
subtle feature patterns, and compares them with trained neural network weights to classify images as

authentic or manipulated. VISTAS uses the MobileNetV2 architecture, a high-efficiency Convolu-
tional Neural Network (CNN) optimized for real-time performance on resourceconstrained devices.

The model is trained on a large dataset of authentic and synthetic facial images, enabling it to detect
microscopic artifacts, hidden noise distributions, and structural inconsistencies left by AI-generated

content. Implemented with TensorFlow and OpenCV, the system provides a scalable and user-friend-
ly solution for media organizations and cybersecurity agencies. By automating deepfake detection,

VISTAS enhances digital trust, supports content verification, and mitigates the risks associated with
AI-generated media in the modern digital landscape.