MUMBAI, India, Aug. 10 -- Intellectual Property India has published a patent application (202541005901 A) filed by S. Brindha on January 24, 2025, for "adversarial Machine Learning For Deepfake Detection And Generation".
Inventors include S. Brindha; K. Sudha; I. N. Sountharia; V. S. Thaneshvar; R. Rathivarman; J. S. Jayanishanth; R. Udaya Ganesh; T. G. Mouriyan; K. L. Vishal; M. Sidharth; and G. Aathish Kumar.
The application for the patent was published on July 31, 2026, under issue no. 31/2026.
Abstract: The burgeoning field of deepfake technology has made adversarial machine learning (AML) necessary for both the creation of manipulated media and its detection. Deepfake technology enables the creation of hyper-realistic but artificially generated images, videos, and audio, often used in entertainment, education, and virtual reality. However, its misuse for misinformation, fraud, and identity theft has raised significant concerns. To address these issues, this project proposes a system leveraging adversarial machine learning techniques to both enhance the realism of deepfakes and improve defenses against malicious usage. It would implement a system that uses UANs tor generating state-of-the-art deepfakes and adversarially train detection models to look out for manipulated content in images and videos. The proposal will fm1her include robust adversarial training of a framework that can automatically evolve with new generative techniques, thus keeping up the effectiveness of the proposed system against evolving threats. To further enhance security of this system, it employs a biometric-based authentication with a ''& multi-layered forensic analysis, which includes the following I. Ad versa rial Perturbations: .Training detection models on adversarially crafted examples in various forms improves the robustness of the detection models against deceptive inputs. 2. 1- - Multi-Modal Forensic Cues: Adversarially robust models analyze inconsistencies in audio- visual synchronization, lighting, and facial movements .The system guarantees both enhanced generation and robust detection by leveraging ..... o something the model learns, something the model detects, and something adversarially :egn generated to create a secure ecosystem for deep fake-related applications. By using these methodologies, the project aims Advance Deepfake Generation: Using state-of-the-art GAN architectures like StyleGAN for generating ultra-realistic media. Improve Detection: Using adversarially trained classifiers to identify forgery patterns 111 manipulated media. Address Ethical Concerns: Setting guidelines and tools for prevention of malicious misuse, encouraging constructive applications of deepfake technology. This project increases the trustworthiness of deep fake technology for constructive application while being secure against the risks. It is contributing toward secure and ethical advancement of AI in media synthesis. The rapid growth of deepfake technology calls for advanced adversarial machine learning (AML) techniques to improve both the creation of realistic media and its detection. This invention introduces a system that uses AM L to not only enhance deep fake generation but also strengthen defenses against its misuse. By using generative adversarial networks (GANs) like StyleGAN, the system can create highly realistic images, videos, and audio for applications in entertainment, education, and virtual reality. To prevent misuse, such as spreading false information or committing fraud, the system includes adversarial training to make detection models more robust. These models are designed to spot manipulations by identifying inconsistencies in elements like audio-visual -synchronization, lighting, and facial movements. The system also adds an extra layer of security with biometric authentication and multilayered forensic analysis. Its adaptive design evolves alongside new generative technologies, ensuring it stays effective against future threats. Key features include training models with adversarial examples to handle deceptive inputs, using multi-modal analysis to catch subtle manipulation, and a self- updating architecture for long-term reliability. This· invention aims to push the boundaries of deep fake creation with advanced GANs, improve detection with smarter classifiers, and encourage ethical use by offering tools and guidelines to prevent abuse. By combining advanced creation, strong detection, and ethical safeguards, this system provides a secure and reliable way to use deepfake technology responsibly while promoting innovation in Al-powered media.
Disclaimer: Curated by HT Syndication.
