Implementation of Efficient Deep Fake Detection Technique on Videos Dataset Using Deep Learning Method

Abstract

Deep fake technology has recently made tremendous advances that have made it possible to produce incredibly realistic fake audio, video, and image media. These materials present serious difficulties for people. Impersonation, false information, or even a national security danger, could compromise authentication. There is currently an arms race between deep fake creators and deep fake detectors as a result of the several deep fake detection algorithms that have been suggested to keep up with these rapid advancements. But these detectors are typically unreliable and frequently miss deep fakes. This study emphasizes to suggest a machine learning technique to minimize the difficulties they encounter in identifying deep fakes in videos dataset. DL and ML is used in this study to proposed a neural network for detection deep fakes from the videos. This study initially selects a video dataset from the from the well-known Kaggle dataset repository. Secondly this dataset is augmented into two classes, real videos and fake videos, and the dataset is divided into training and testing. Thirdly, the preprocessing of dataset is done by face-extraction, region of interest selection and frames extraction to detect the real and fake videos. Fourthly, Neural Network is applied on the processed dataset and evaluate the model by calculated the accuracy. Finally, Proposed model is compare with the other models such as (Resents, Inception V3 and vision transformers). The comparison shows that our proposed model is perform well on the processed dataset as compared to other models and achieved accuracy of 94.86 percent.