DeepFake Face Detection using Machine Learning with LSTM
Thipparthi Vignesh, Potharlanka Harish Tarun, Ryagalla Parthav, V Bhargavi · 2024
Fake face images that are increasingly convincing and realistic can be created because to the development of face image manipulation (FIM) technologies like Face to Face and Deepfake, which can damage the legitimacy and trustworthiness of online content. Malicious uses of these technology include blackmailing people, posing as celebrities, and disseminating false information. As a result, creating trustworthy and strong techniques to identify FIM and safeguard the integrity of digital media is essential. Numerous current techniques utilize on models built on convolutional neural networks (CNNs), which are capable of detecting FIM by examining a face’s visual characteristics. But because these models are frequently tested and trained on certain datasets or circumstances. Furthermore, they might not be able to record the temporal information that is included in video data and can be used to identify irregularities or strange anomalies in FIM videos. We provide a novel method that uses both geographical and temporal information to detect FIM in order to get over these difficulties. We present a new type of residual network called CRNet, which is dependent on Convolutional Long Short-Term Memory (LSTM) and is capable of processing a series of consecutive pictures taken from a movie. The model can learn temporal information because to its design, which is essential for spotting oddities that occur in between frames of FIM movies. We performed extensive tests with several kinds of FIM videos from the Kaggle dataset.