DeepFake Detection using 3D-Xception Net with Discrete Fourier Transformation

Adeep Biswas,Anil Kumar Kakelli,Debayan Bhattacharya

doi:10.52547/jist.9.35.161

Abstract

The videos are more popular for sharing content on social media to capture the audience’s attention. The artificial manipulation of videos is growing rapidly to make the videos flashy and interesting but they can easily misuse to spread false information on social media platforms. Deep Fake is a problematic method for the manipulation of videos in which artificial components are added to the video using emerging deep learning techniques. Due to the increase in the accuracy of deep fake generation methods, artificially created videos are no longer detectable and pose a major threat to social media users. To address this growing problem, we have proposed a new method for detecting deep fake videos using 3D Inflated Xception Net with Discrete Fourier Transformation. Xception Net was originally designed for application on 2D images only. The proposed method is the first attempt to use a 3D Xception Net for categorizing video-based data. The advantage of the proposed method is, it works on the whole video rather than the subset of frames while categorizing. Our proposed model was tested on the popular dataset Celeb-DF and achieved better accuracy.

Highlights

In recent times, the usage of videos has increased rapidly for different purposes such as marketing, news, and entertainment [1]
The rest of the paper is arranged in the following mannersection 3 discusses the relevant research work which has already been done in this domain, section 4 illustrates details of the deepfake detection model proposed in this paper, section 5 specifies the configuration of our experimental setup and the results that were achieved by the proposed model and in section 6 we provide the conclusion and discuss the further scope for research work in this domain in the future
The experiments which are conducted to determine the performance of our proposed methodology are discussed in detail. This involves the overall experimental setup and resources used to achieve our results, the analysis of the outcome of the testing done on our proposed methodology as well as our model’s performance comparison to other deepfake detection algorithms

Summary

1- Introduction

The usage of videos has increased rapidly for different purposes such as marketing, news, and entertainment [1]. The proposed work has proposed a new deepfake detection technique-3D Inflated Xception Net with Discrete Fourier Transformation which was able to achieve state-of-the-art accuracy results. The main drawback of such an approach is if the right video frame is not selected for the input, the video may get incorrectly categorized since artificial manipulations don't need to be done to all frames of the video when a deepfake is generated This problem has been addressed in our proposed model by converting the 2dimensional architecture of Xception net into 3dimensions and initializing the network by pre-training it on static videos generated from a subset of images of the ImageNet dataset [25]. The rest of the paper is arranged in the following mannersection 3 discusses the relevant research work which has already been done in this domain, section 4 illustrates details of the deepfake detection model proposed in this paper, section 5 specifies the configuration of our experimental setup and the results that were achieved by the proposed model and in section 6 we provide the conclusion and discuss the further scope for research work in this domain in the future

2- Related Work

3- Proposed Methodology

4- Experimental Results

4-1- Experimental Setup

4-2- Results and Analysis

5- Conclusion and Future Scope