Generated Video Detection Method Based On Ambient Signal

In recent years,the continuous upgrading of computer hardware,the rapid development of deep learning technology and the wide application of multimedia digital technology provide sufficient computing power,practical theoretical basis and rich training data for the automatic generation of deep fake video.It only needs a few simple steps to pass the pre trained neural network model,Automatically replace the face in the video with another person.Since the first deep forgery video was released in reddit at the end of 2017,a large number of celebrity spoof videos and political fake news appeared on the Internet,which had a very bad impact on the real world.Therefore,more and more researchers began to explore the automatic generation of video detection technology.At present,the mainstream generation video detection technology is based on deep learning network,through the collection or production of a large number of false video,design a variety of neural networks to identify the false traces in the generated video,but blindly using deep learning detector in capturing false content will not always be effective,because the generation model is based on the same principle of continuous training Enhanced to produce more and more realistic video.In the process of video recording,it is inevitable to capture a variety of Ambient signals.For example,in the environment of AC electric field or AC electrical equipment,the camera will capture the fluctuation of electric network frequency(enf)signal at the same time.This paper considers that the Ambient signal hidden in the face video can be used as an effective information to identify the authenticity of the video,Because Ambient signal has good randomness and naturalness,at the same time,the current deepfake deep learning model does not consider the influence of Ambient signal in training.Therefore,this paper also proposes a video detection method based on unknown Ambient signal.Firstly,the luminance signal of face and background is extracted from the video,and then the spectral features are further extracted,Finally,using machine learning classifier to detect the authenticity of video.The method was evaluated on the experimental data set collected by myself,the detection accuracy of single frame is 86.62%,and the detection accuracy of multi frame fusion in the same second is 98.68%.At the same time,inspired by the natural Ambient signal,this paper also proposes a video detection method based on the artificial Ambient signal.By injecting the artificial Ambient signal into the video in advance,the correlation coefficient between the Ambient signal in the video and the original signal can be directly calculated during the detection,and the authenticity of the video can be accurately determined.Finally,this paper explores how much interference received by the Ambient signal can still be used as a reference signal to detect the authenticity of the video.