SMART DEEPFAKE DETECTION SYSTEM USING CNN FOR DIGITAL IMAGE AUTHENTICITY

Abstract: The rapid development of generative artificial intelligence has increased the availability of realistic synthetic and manipulated facial images. This paper presents a Smart Deepfake Detection System Using Convolutional Neural Network (CNN) for automated binary classification of facial images into real and fake classes. The work extends the supplied project manuscript by providing a fuller IEEE-style treatment of the problem, preprocessing pipeline, CNN architecture, web deployment, evaluation methodology, security considerations, and reproducibility requirements. The implemented system resizes and normalizes an uploaded image, performs CNN inference, and displays the predicted class and confidence through a Flask-based interface. The supplied project evidence demonstrates two functional cases: a representative genuine face is returned as “Real Face Detected,” while a representative manipulated/synthetic face is returned as “Fake Face Detected.” Because the source material does not provide dataset counts or statistical metrics, this paper deliberately does not fabricate accuracy, precision, recall, F1-score, ROC-AUC, or confusion-matrix values. Instead, it specifies a publication-ready evaluation protocol that can be populated from the actual experimental run.

Keywords: Deepfake detection, convolutional neural network, facial image forensics, synthetic media, image classification, Flask, digital security, deep learning.