Deep Fake Detection Based on Hybrid Approach Involving Residual Network Deep Learning Architecture

This Deep fake Technology aims at generating synthetic video content or morphed video content which is very realistic to the naked eye. It involves the use of deep learning algorithms in the domain of video processing. In most cases, such artificially generated video contents are very difficult to get differentiated from naturally captured video signals. The existing methodologies of detection involved image tamper detection and digital forensics algorithms. These algorithms are ineffective in detecting videos tampered using deep fake technology. The proposed technique involves training and building of deep learning models that works like a black box to detect the artefacts crucial in the detection of video tampering. This paper presents an overview of the technology advancements in the field of deep fake video detection techniques over recent years. This paper also performs a comparative analysis on the performance of hybrid algorithms involving residual network deep learning architecture in the detection of video signals morphed using deep fake technology. The training datasets are obtained from Face forensics to build the model. The algorithm is augmented with image quality assessment, temporal consistency analysis and facial landmarks analysis to improve the detection algorithm. The performance metrics involve the generation of confusion matrix and generation of accuracy measures. The results provide a promising direction towards the construction of better detection models to safeguard against current and future threats.