Real-Time Deepfake Detection System Using Custom CNN and OpenCV: A Web-Based Approach for Educational and Research Applications

Abstract

As the generation of synthetic media grows, more advanced, and able to tell real content from fake content. Artificial intelligence for producing what have been called deep- fakes poses growing challenges to digital security and media verification. We propose a novel real-time detection model incorporating custom neural network architectures using modern web technologies. Our monoxide, CFD and LES-CNN approach- integrating a specially established Convolutional Neural Network (CNN). Principal detection engine having a network forming part, further including OpenCV Haar Cascade algorithms for face detection. All embedded into a React-Flask Web platform. Through training with a carefully created dataset of 6,000 heterogeneous images, our system gets the classification accuracy of 85% when discriminating authentic from artificially modified face images utilizing key CPU processing. Primary innovations encompass: (1) a simplified dual- architecture design with our custom CNN with a powerful backup detection mechanism, (2) extractor facial recognition using OpenCV Haar cascades,(3) a user-friendly web-based user interface supporting both static image and video content analysis, and (4) in-depth evaluation of all available synthetic generation techniques including transformation with Style GAN, Progressive GAN, and Face App. Our solution is especially insightful for cyber security educational purposes, security initiatives, content authentication tasks and academic study performance taking into consideration the highest possible performance on conventional computing systems without specialized GPU devices.