A deep fake detection approach for cyber security threat based on deep learning and diffusion-osmosis model

Abstract

Advancements in the realm of digital image alteration have resulted in the widespread adoption of deep fake technology, presenting notable obstacles to the field of digital forensics and cyber security. This article suggests an original methodology that combines hybrid anisotropic diffusion-variational osmosis for image filtration, U-Net for segmentation and edge detection, and a hybrid graph convolutional network (GCN) employing separable convolution for categorization in deep fake detection. The amalgamated filtration technique improves image fidelity while conserving crucial particulars, facilitating accurate segmenting and edge detection essential for recognizing altered areas. The GCN framework utilizes graph-oriented learning to extract intricate features, supported by effective separable convolutions for precise categorization of genuine and fake images. The suggested approach is assessed using extensive datasets, showcasing superior performance in detecting deep fake images compared to conventional techniques. Besides technological advancements, this study highlights the broader repercussions of deep fake detection in cyber security. Deep fakes, adept at duping automated systems and human perception, present substantial risks encompassing misinformation, identity theft, financial deception, and breaches in national security. Efficient detection techniques, such as those outlined here, play a crucial role in mitigating these dangers and upholding digital trust and integrity. The proposed methodology is positioned within the wider framework of cyber security, accentuating its significance in combating the escalating threats and offenses linked to deep fake technology. It accentuates the technical progressions while emphasizing the vital necessity of robust identification mechanisms in tackling cyber security challenges arising from digital alterations. The empirical results illustrate that the suggested technique outperforms other deep fake image detectors and achieves a remarkable accuracy rate of 99.71% in the dataset utilized in this study.