Deepfake Detection in Media Files - Audios, Images and Videos

摘要

Recent advancement in deep learning has applied to solve various complex problems ranging from big data analytic to computer vision and human-level control. One among them is the deepfake technology which becomes a real threat to privacy, democracy, and national security. Deepfake is hyper-realistic digitally manipulated videos to depict people saying and doing things that never actually happened. This technology has been used in many fields in film industries for recreating videos without re-shooting, awareness video generation such as creating voices of those who have lost theirs or updating episodes of movies without re-shooting them at very low cost. This technology has many harmful usages in social media, pornographic sites, etc. to deface peoples which largely dominate the positive side of this application of deep learning. Also, the creation and spreading of these videos are increasing rapidly along all fields of media files. Therefore, it is very much important to develop efficient tools that can automatically detect the deepfake in these videos and thus reduce the public harm caused by such videos. In the early stages of deepfake detection, traditional technologies like signal processing, image processing, lip-syncing, etc were used but this provides very little accuracy when combined with the recent technologies of deep learning. So, here a system is proposed that can automatically detect the deepfake in media files such as images, videos, and audios. This uses an image processing approaches combined with deep learning which detects the inconsistency that exists in fake media.