Face Warping Detection and Localization in a Digital Video using Transfer Learning Approach

Abstract Artificial Intelligence (AI) is making its entire existence safer in this digital world. Another deepfake technique, face warping, is booming in the media. Digitally altered videos of a person’s face or body are considered as deepfakes. Face warping is the process of digitally processing a face to significantly distort shapes represented in itself. Warping can be used to correct image distortion or for creative purposes such as morphing. It is important to track such face warping in images and videos so that the very purpose of face warping is not used for destructive purposes. The deep learning approaches are used to determine and localize face warping in videos. In the proposed approach, the input video is extracted to perform various image preprocessing techniques that refine the video into a format that is more likely to classify the classes efficiently. Transfer learning is employed and the pretrained model is adopted to train using Convolutional Neural Network (CNN) with the source videos to detect and localize face warping. It is perceived that the proposed model is able to detect and localize the warped areas in the face satisfactorily with an accuracy of 89.25%.