Research On Copy-move Forgery Detection For Digital Videos Based On Deep Learning

With the rapid development of multimedia technology and mobile electronic devices,various digital videos have recently brought a lot of fun and convenience to people,which become indispensable in our daily life.However,some malicious persons have forged or will forge digital videos for their own benefits,which results in numerous information security issues.Therefore,effective detection of integrity,authenticity and originality of digital videos is an important guarantee in the field of video forensics.However,the existing digital video forgery detection methods suffer from insufficient feature extraction capability and low detection accuracy.With the great success of deep learning in image recognition and video classification,researchers have gradually applied deep learning to video forgery detection.After the introduction of relevant theories on video forensics,this thesis focuses on copy-move forgery detection for digital videos based on deep learning.The main work is described as follows.(1)A video forgery detection method is proposed based on convolutional auto-encoders,which is constructed by combining a convolutional neural network with an auto-encoder.The parameters of convolutional kernels are configured to extract features,which are based on video characteristics.To improve the detection performance,a cascaded high-pass filter bank is designed to filter video data,which is helpful for highlighting forgery and for feature extraction.The experiment results show that the proposed method is superior in detection accuracy compared with the traditional detection methods.(2)A deep learning method for video forgery detection is proposed based on global-and-local feature fusion.In order to solve the problems of low detection accuracy and insufficient feature extraction capabilities in existed methods,the proposed method introduces multi-network feature fusion,which is implemented by combining global feature extraction subnetwork and local feature extraction subnetwork.The subnetworks are to respectively extract the global and local features from the videos.Global and local features are fused in feature fusion layer.The comparison experimental results show that,compared with the existing video forgery detection methods,the proposed method has a stronger ability of feature extraction and higher forgery detection accuracy for videos.