abstract

Deepfake technology poses significant threats to security, privacy, and trust in digital media. This paper introduces a novel approach to deepfake detection by analyzing pixel patterns related to eye blinking dynamics in real-time video streams. By leveraging machine learning algorithms to detect anomalies in eye blinking, our method offers an effective and efficient solution for identifying deepfake content. This study, focused on the Indian context, provides insights into the implementation challenges, accuracy, and potential applications of this technology.

Thanks to advances in processing power, deep learning algorithms have made it very simple to create extremely lifelike synthetic movies, or "deep fakes." These movies provide serious threats including extortion, political manipulation, and staging phoney terrorist incidents since they can realistically switch faces. In this research, a unique deep learning approach for effectively differentiating between real movies and AI-manipulated ones is presented. The suggested approach extracts frame-level features from movies using a Res-Next Convolutional Neural Network (CNN), picking up on minute details and patterns in each frame. A recurrent neural network (RNN) built on Long Short-Term Memory (LSTM) is then trained using these characteristics. The LSTM network uses its capacity to record temporal information to evaluate the series of frames and detect whether the video has been changed.

The method is extensively evaluated on a sizable, well-balanced, and diverse dataset that was produced by fusing together many pre-existing datasets, including Face Forensics++, the Deep Fake Detection Challenge, and Celeb-DF. This extensive dataset improves the model's ability to detect deep fakes in real-world settings by simulating real-time events. The system attempts to provide strong detection skills by including these datasets in a way that reflects a variety of video quality and processing approaches. The ultimate goal is to utilise AI to counter the risks that AI poses by developing a trustworthy technique for automatically identifying deepfakes. With its use of state-of-the-art deep learning algorithms, this methodology marks a major breakthrough in the battle against digital manipulation and safeguards the integrity of video footage.