Research On Key Technologies Of Passive Forensics For Multimedia

With the development of Internet technology,massive multimedia information can be quickly and widely disseminated by means of social networks.This provides a better way for people to understand information,but also leaves an opportunity for illegal forgers to disseminate false information.Therefore,how to identify the authenticity of multimedia content in the massive multimedia data has become a hot and difficult issue.We grasp these changing trends in the field of multimedia information security,and summarize the research status nowadays.We take the image resampling operation,image median filtering operation,image inpainting operation,video object removal operation as the research objects.The main contributions of our work are as follows:1)image resampling detection based on blind deconvolutionBy the technology of blind deconvolution,an image resampling detection method is proposed.The proposed method uses blind deconvolution to restore image editing history,and can effectively detect image resampling operation and identify interpolation operation types.This method can be widely used in resampled images after zero-order interpolation,second-order interpolation and third-order interpolation,and can accurately detect non-periodic interpolation and periodic interpolation.Experimental results show that the method is robust to JPEG compression and Gaussian noise.2)image median filtering detection based on generative adversarial networkBy the recent technology of generative adversarial network,a detection framework for median filtering detection is proposed.We extend median filter detection from grayscale to RGB images.Prior knowledge in natural RGB images is applied to median filter detection.Using the original images to train the generative adversarial network model,the median filtered images in the test process will break the balance between the generator and the discriminator in the training process,thus the tampered images can be discriminated.The experimental results show that the method has competitive detection results for small image blocks and JPEG compression.3)sparsity-based image inpainting detection based on canonical correlation analysis with low-rank constraintsTo fight against the problem of sparsity-based image inpainting,a detection method based on canonical correlation analysis with low-rank constraints is proposed.Through mathematical analysis and preliminary experiments,the latent connection between sparsitybased inpainting and traditional canonical correlation analysis is revealed.In order to characterize the inpainted signals correctly,an improved Canonical Correlation Analysis(CCA)expression is proposed by adding low-rank constraints,which can better characterize the correlation between signals and distinguish the original images from inpainted images.Experimental results show that the proposed method can detect sparsity-based image inpainting operations better compared with other methods.4)video forensics based on sparse representationAiming at the problem of copyright protection,a general solution of TV logo removal detection is proposed.Firstly,sparse learning is used to construct a low-dimensional feature set,which can effectively distinguish tampered regions from original regions.At the same time,a video tampering localization algorithm is designed by exploring the temporal and spatial prior knowledge from a novel perspective.It employs the improved three-dimensional morphological operation to improve the detection results.In total,four effective multimedia forensics algorithms are proposed.The prior information in forensic tasks is mined,and the differences between original signals and tampered signals are revealed,which enriches the research contents in this field.