Desing And Implementation Of Fake Face Image Detection Model Based On Image Forensics Technology

With the rapid development of the media,nowadays,the majority of the people understand the news and social events by means of digital media,especially video.At the same time,people often default that the content of news reports is honest and trustworthy,which provides an opportunity for lawbreakers to modify the face image in the video to guide people to understand the meaning expressed in the original video.Therefore,it is of great significance to distinguish whether the video is real or not for the government public opinion management and event management and control.At present,the research of fake face detection can be divided into two categories: one is based on the specific and significant images in the image,such as eyes and nose,but the counterfeiter can be trained to deceive the audience again,which has the disadvantage of being attacked again;The second method is based on the low-level semantic features,which directly puts the face region into the classification model for low-level semantic feature extraction and image classification for detection.This kind of method is not affected by the shortcomings of the previous method,but both methods only use the face region in the image,It can not effectively use the information of other areas of the image or the operation taken in the process of counterfeiting for auxiliary detection.In this paper,based on the analysis of the underlying semantics,background noise information and detection interpolation operation are added to further improve the detection accuracy on the basis of the original classification model.In this paper,the fake face image in video media platform is taken as the research scene.At the same time,considering the underlying semantic information and background noise information,a Multi Strategy detection model and algorithm is proposed to detect fake face image more accurately and comprehensively.The main contents of this paper are as followsFirstly,this paper proposes a method to extract the noise of face and background region.Based on the photo response non-uniformity algorithm and the self attention mechanism,a multi-scale background noise extraction algorithm is designed firstly,and then a detection method is proposed by using the self attention mechanism to fuse these different scale noise features.Secondly,a method of fusing the interpolation operation and the underlying semantics in the process of fraud detection is proposed.Based on the rich steganalysis model,the previously ignored interpolation operation is included in the detection module.Firstly,the image is put into the steganalysis model to extract the interpolation matrix with interpolation features.Then,the interpolation matrix and the original image are extracted by residual convolution network and Xception Net.The two results are spliced to obtain the feature vector with interpolation and underlying features,and the detection method is based on this.Thirdly,based on the above two methods,a fake face image detection model is established by fusing the three kinds of information.Based on the existing data sets,the effectiveness of the algorithm is verified by experiments,and compared with the relevant research results.Through experimental comparison:compared with the latest research,the ACC value of the proposed fake face detection model is improved by3.6%.The F1 value is increased by 3.8%,which indicates that this method is more suitable for fake face detection.Finally,based on the fake face detection model of the third point,the paper designs and implements the fake face detection prototype system,and shows it.The research background of this paper is based on the school enterprise cooperation project: social network information analysis and image technology research and consultation.In addition to better detection results than previous research methods,this detection method can also be used to help the social media platform itself improve the quality of content,improve the user's real experience,and ultimately create a better image and video trusted browsing environment for users.