DeepScan Sentinel: Multimodal Deepfake Detection & Cybersecurity Platform

Abstract

In the past few years there is vast increase in the technology and the usage of modern technologies which incorporate artificial intelligence were reaching its peak among the peoples. As the technology advances, the possibilities of involving digital crimes were one of the considerable issues and needed to mitigate in various method. One of the issue that related to it was deepfakes (DFs) which involves in morphing the faces, shape, background, voices, vocal lines, sounds and texts. This paper presents the project “DeepScan Sentinel: Multimodal Deepfake Detection & Cybersecurity Platform” to identify the deepfake in the digital media like images, videos, audios and texts in a single platform by integrating the artificial intelligence and machine learning for its identification. It also consists of various other cybersecurity tools to practice and using in real time usage which helps to learning cybersecurity for beginners. By using this platform we can quickly and effectively identify the morphed part of the digital data. In other platforms there is no combined method of identification to identify the deepfakes in digital media. But this platform had image, videos, audio and texts morphing identification methods together in one screen, which will help the users in more efficient way to identify the deepfakes in digital media. And apart from that it had some awareness contents, that could help the users to get some knowledge about the deepfakes, how it was created and further how we can identify that.

Introduction

Summary

Deepfakes have emerged as a significant technological threat, capable of manipulating images, audio, video, and text using AI and machine learning. These forgeries pose risks to individuals, organizations, and society through identity theft, misinformation, and unauthorized content creation. As a response, the research paper “SLM-DFS: A Systematic Literature Map of Deepfake Spread on Social Media” reviews 286 studies (2018–June 2024), focusing on the spread of deepfakes via social media and the need for effective, scalable detection solutions.

Existing Research Areas

1. Image Morphing: Includes facial manipulation and morphing attacks that compromise biometric systems. Various detection methods, including GAN-based and forensic approaches, are explored to counter image tampering.

2. Video Morphing: Builds upon image morphing, adding complexity through temporal manipulations. Datasets like FaceForensics and models using spatio-temporal analysis help improve detection accuracy.

3. Audio Morphing: Involves changing voice or sound characteristics. Techniques use deep learning (e.g., CNNs with MFCCs) to detect synthetic audio, crucial in biometric and forensic contexts.

4. Text Morphing: Less common but involves gradual textual transitions or manipulations. It’s studied in contexts like natural language inference (NLI) and disinformation detection.

Proposed System Overview

A multimodal deepfake detection system is proposed, aiming for holistic and real-world effectiveness. It integrates image, audio, video, and text analysis with network-level security tools to offer a robust and scalable solution.

A. Data Pipeline

• Image: Uses MTCNN for face detection, normalization.

• Audio: Converts audio to mel-spectrograms; analyzed via WavLM and SyncNet.

• Video: Extracts frames and features for TimeSformer and XCeption analysis.

• Metadata: Analyzed with DeepSeek R1 to flag anomalies.

B. Model Ensemble

• Image: ViT + XCeption for pixel-level and spatial manipulations.

• Video: TimeSformer (temporal) + XCeption (spatial anomalies).

• Audio: WavLM + SyncNet for voice and lip sync inconsistencies.

• Text: DeepSeek R1 for analyzing embedded text/metadata.

C. Cross-Modal Fusion

• Features from all media types are aligned using cross-modal attention, allowing the system to detect inconsistencies across modalities and provide stronger validation of media authenticity.

D. Security Integration

• Network tools (e.g., Nmap, Netcat, HackerTarget API) add a layer of cybersecurity, scanning IPs and ports linked to suspicious media to detect vulnerabilities and assess threats.

Conclusion

When compared to similar deepfake detection projects, this system offers a more comprehensive and robust solution by integrating multimodal analysis, advanced models, and security tools. While many existing systems focus on a single modality such as video or audio, this project stands out by combining image, audio, video, and text analysis using cutting-edge models like ViT, TimeSformer, WavLM, and DeepSeek R1. The multimodal approach significantly enhances detection accuracy by cross-referencing data inconsistencies across different types of media. Another key differentiator is the integration of security tools such as Nmap and Netcat, which provide an added layer of trust by identifying potential vulnerabilities in media-sharing platforms. While most existing systems lack this feature, it strengthens the practical, real-world applicability of this solution, particularly for platforms concerned with both media authenticity and cybersecurity.In terms of performance, the project is expected to achieve competitive accuracy (>90%) across various datasets and modalities, with faster inference times and greater robustness against adversarial attacks. Unlike many current systems that are limited by dataset size and modality, this system leverages a large dataset of 190,000 images, 6,000 audio clips, and 6,100 videos, ensuring better generalization to real-world scenarios. This project offers a significant advancement over existing deepfake detection systems by providing a multimodal, secure, and scalable solution that addresses both media authenticity and cybersecurity concerns. Its comprehensive approach and ethical design ensure that it not only meets high accuracy standards but also aligns with privacy and legal requirements, positioning it as a publishable and impactful solution in the deepfake detection domain.

References

[1] SLM-DFS: A Systematic Literature Map of Deepfake Spread on Social Media, E.-S. Atlam, M. Wills, M. Selman, M. Bedair, 2025. [2] Handbook of Digital Face Manipulation and Detection, Christian Rathgeb, Ruben Tolosana, Ruben Vera-Rodriguez, Christoph Busch, 2022. [3] FaceForensics: A Large-scale Video Dataset for Forgery Detection in Human Faces, Andreas Rössler, Davide Cozzolino, Luisa Verdoliva, Christian Riess, Justus Thies, Matthias Nießner, 2019. [4] Towards a Robust Framework for Multimodal Hate Detection: A Study on Video vs. Image-based Content, Girish A. Koushik, Diptesh Kanojia, Helen Treharne, 2025. [5] Deepfake Detection over Different Media Types Using Deep Learning Algorithms, I. Amerini, L. Ballan, R. Caldelli, A. Del Bimbo, G. Serra, 2023 [6] DeepFake Detection for Human Face Images and Videos: A Surve, Ping Liu, Xin Shu, Zicheng Liu, Changsheng Xu, 2021. [7] Face Morphing, a Modern Threat to Border Security: Recent Advances and Open Challenges, Johannes Merkle, Ralph Breithaupt, Christoph Busch, 2019. [8] GAN-Generated Faces Detection: A Survey and New Perspectives, Richa Singh, Mayank Vatsa, Nalini K. Ratha, 2022. [9] Video Detection Method Based on Temporal and Spatial Foundations for Accurate Verification of Authenticity, I. Amerini, R. Caldelli, L. Ballan, A. Del Bimbo, G. Serra, 2022. [10] Identifying and Minimizing the Impact of Fake Visual Media: Current and Future Directions, Andreas Savakis, Chen Feng, 2024. [11] Text Morphing, Shaohan Huang, Yu Wu, Furu Wei, Ming Zhou, 2018. [12] Investigation and Morphing Attack Detection Techniques in Multimedia, Ateeq Ur Rehman, Muniba Ashfaq, Zahid Wadud, 2023. [13] Image Morphing Concept for Secure Transmission of Image Data, Anant M. Bagade, S.N. Talbar. [14] Image and Video Forensics, Andreas Uhl, Stefano Dragoni, 2021. [15] Analytics and Applications of Audio and Image Sensing Techniques, Frederic Y. Shen, Daniel S. Hamlin, 2020. [16] Image and Video Manipulation: The Generation of Deepfakes, Jun-Yan Zhu, Richard Zhang, Phillip Isola, Alexei A. Efros, 2017. [17] An rTMS Study into Self-Face Recognition Using Video-Morphing Techniques, W. Jiang, J. Zhang, S. Wang, R. Luo, 2019. [18] Deepfake Forensics: A Survey of Digital Forensic Methods for Multimodal Deepfake Identification on Social Media, Samay Pashine, Sagar Mandiya, Praveen Gupta, 2021 [19] A Comprehensive Review of Face Morph Generation and Detection of Fraudulent Identities, M. Hamza, S. Tehsin, M. Humayun, 2022. [20] Face Morphing Attack Generation and Detection: A Comprehensive Survey, Raghavendra Ramachandra, Kiran Raja, Christoph Busch, 2021. [21] Face Recognition Systems Under Morphing Attacks: A Survey, Johannes Merkle, Ralph Breithaupt, Christoph Busch, 2019. [22] Automatic Audio Morphing on Detached Sound Waveforms, Amarjot Singh, K. Anishya Sruthi, 2012. [23] Statistical Analysis & Reliability Testing of Various Acoustic Elements in Forensic Examination of Morphed Voice Samples, M. Rashid, P. Smith, 2022. [24] High-Level Audio Morphing Strategies, Wesley Hatch. [25] Developing a Timbre Morphing Package: The Process of Automatic Feature Identification, Ciaran Hope, 2012. [26] Sound Morphing Strategies Based on Alterations of Time-Frequency Representations by Gabor Multipliers, Anaïk Olivero, Philippe Depalle, Bruno Torresani, Richard Kronland-Martinet, 2012. [27] Analyzing and Improving the Image Quality of StyleGAN, Tero Karras, Samuli Laine, Timo Aila, 2020. [28] MorphNLI: A Stepwise Approach to Natural Language Inference Using Text Morphing, Vlad-Andrei Negru, Robert Vacareanu, Camelia Lemnaru, Mihai Surdeanu, Rodica Potolea, 2025.

Copyright

Copyright © 2025 KamalaKannan R, Arun Selvam M, Mohammed Aslah P S, Vishnu M P, Vishnu Sajikumar. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.