Deepfake Detection Using Hybrid CNN-LSTM Architecture with Blockchain-Based Media Authentication

The rapid advancement of artificial intelligence has enabled the creation of highly realistic deepfake images and videos, posing significant threats to digital media authenticity, cybersecurity, and public trust. Existing deepfake detection techniques often rely on either spatial or temporal information, resulting in limited robustness against sophisticated manipulations. This paper proposes a hybrid CNN-LSTM framework integrated with blockchain-based media authentication for reliable deepfake detection and secure media verification. The Convolutional Neural Network (CNN) extracts discriminative spatial features from facial frames, while the Long Short-Term Memory (LSTM) network captures temporal inconsistencies across video sequences. To ensure media integrity, SHA-256 hashing is combined with blockchain technology to provide tamper-proof authentication and provenance tracking. The proposed model is evaluated using FaceForensics++, DFDC, and Celeb-DF datasets. Experimental results demonstrate high detection accuracy, improved robustness, and secure verification compared with existing approaches. The proposed framework offers an efficient solution for social media platforms, journalism, digital forensics, and cybersecurity applications.