MUMBAI, India, June 24 -- Intellectual Property India has published a patent application (202441097255 A) filed by Dhaksata S; Chandru R; Niranjan S; and Pandiyan M on December 10, 2024, for Deep Fake Face Detection Using Deep Learning.

Inventors include Dhaksata S; Chandru R; Niranjan S; and Pandiyan M.

The application for the patent was published on June 12, 2026, under issue no. 24/2026.

Abstract: The rapid advancement of deep fake technology poses significant challenges for digital content verification and security. This study evaluates the effectiveness of multiple machine learning models for deep fake detection, including K-Nearest Neighbors (KNN), Random Forest, an Ensemble Model, and a Multi-Layer Perceptron (MLP) neural network. Using an image dataset from Kaggle, the system was trained and tested within a cloud-based Google Colab environment, employing TensorFlow/ Keras for deep learning and Scikit-leam for machine learning. Among the models .tested, the MLP neural network exhibited superior performance with a mean cross-validation accuracy of 73.2%, indicating its capacity to generalize effectively to new, unseen data. The hierarchical feature extraction capability of MLPs played a significant role in distinguishing real faces from deep fakes, contributing to improved detection efficacy. In contrast, simpler models like KNN and Random Forest demonstrated lower recall for deep fake detection, suggesting that, while they maintained reasonable precision in recognizing real faces, they missed a substantial number of deep fakes. This highlights a critical limitation in their ability to detect manipulated content reliably. The study underscores the importance of further enhancing model architectures and incorporating strategies such as data augmentation, hybrid models, and transfer learning to address recall issues. These improvements are essential to build more robust and effective deep fake detection systems, especially in high-stakes scenarios where accuracy and reliability are paramount.

Disclaimer: Curated by HT Syndication.