International Journal of Science, Engineering and TechnologyAuthors: Research Scholar Priya Mishra, Dr. Deepika Pathak
Abstract: The rapid advancement of artificial intelligence and generative models has significantly increased the creation and dissemination of deepfake images, posing severe threats to digital security, media authenticity and public trust. Conventional deepfake detection methods often suffer from limited generalisation capability, high computational complexity and reduced performance against advanced synthetic image generation techniques. This review paper presents a comprehensive analysis of hybrid deep learning frameworks for robust and computationally efficient image-based deepfake detection. The study critically examines existing architectures including Convolutional Neural Networks (CNNs), Vision Transformers (ViTs), attention-based models and explainable artificial intelligence (XAI)-enabled frameworks, and investigates the role of preprocessing and feature enhancement techniques such as normalisation, edge detection, artifact analysis and feature fusion in improving detection accuracy. Eleven studies are reviewed and organised into three themes covering deep learning architectures, preprocessing and feature enhancement, and benchmarking with societal impact. The reviewed evidence is consolidated into comparative tables that map each study to its focus, approach, principal finding and limitation, alongside a comparison of architecture families and a summary of preprocessing techniques. The review identifies six research gaps related to computational efficiency, cross-dataset generalisation, model interpretability, preprocessing strategy, benchmark inconsistency and lightweight hybrid design, and maps each to a corresponding future research direction. Directions focusing on lightweight hybrid architectures, explainable frameworks, benchmark-driven evaluation and multimodal detection strategies are presented to support the development of reliable and scalable deepfake detection systems.
