MUMBAI, India, Sept. 22 -- Intellectual Property India has published a patent application (202641109852 A) filed by Vallurupalli Nageswara Rao Vignana Jyothi Institute Of Engineering And Technology on September 13, 2026, for An Artificial Intelligence-Based Deepfake Face Detection System Using Deep Learning-Based Facial Feature Analysis.
Inventor includes Mrs. V. Manjula.
The application for the patent was published on September 18, 2026, under issue no. 38/2026.
Abstract: ABSTRACT [0036] A deep-learning system for distinguishing authentic videos from artificially generated deepfake videos is disclosed. Frame-level visual features are extracted by a ResNeXt convolutional neural network and subsequently processed by a long short-term memory recurrent network that classifies the entire sequence as real or manipulated. Training employs a balanced mixture of publicly available datasets including Celeb-DF, FaceForensics++, and the Deepfake Detection Challenge collection, ensuring robustness across diverse generation methods. Preprocessing isolates facial regions, normalizes frame count to 150, and resizes each face crop to 112 by 112 pixels. The resulting classifier operates on short video segments and returns both a binary decision and a confidence score, enabling practical deployment through a web interface for real-time verification. [0037] The increased processing power has made deep learning algorithms so effective that it is now quite possible to create a video that is clearly human-generated, a phenomenon frequently referred to as a"deep fake." Circumstances wherein these realistic face-swapping deepfakes are utilized to createadult content, stage terrorist attacks, and provoke political turmoil, it‘s easy to envisage people whoare blackmailed. In this paper, we introduce a revolutionary deep learning method that can distinguishreal films from fake ones produced by artificial intelligence (AI). Our method is capable of automatically identifying replacement and replica deep fakes. Using artificial intelligence (AI) against itself is our strategy for defeating it. Our approach extracts the frame-level characteristics using a Res-Next Convolution neural network, and uses these features to learn the Recurrent neural networks(RNNs) based on long short-term memory (LSTM) are used to classify whether or not the video hasbeen altered. Is the video real or deepfake? In order to mimic the actual situations and create the model performs better on real-time data, thus we test our approach on a sizable number of mixed,balanced data sets that were created by combining several existing data sets, such as Celeb-DF, Face-Forensic++, and Deepfake Detection Challenge. We also demonstrate how our systemmay use a very straightforward and reliable method to get competitive results.
Disclaimer: Curated by HT Syndication.
