Research On Algorithms For Improving The Quality Of PRNU Extraction

With the rapid development of mobile imaging technology,people have become accustomed to taking digital images with their phones and sharing them via the internet,greatly enriching the ways in which people communicate in their daily lives.However,the frequent occurrence of cases in which illegal criminals use digital images to spread illegal and criminal content deeply endangers social harmony.Therefore,how to identify the source of the digital image’s capture device has become a hot research topic in the field of information security.Currently,PRNU(Photo-Response Non-Uniformity)noise,as a unique "fingerprint" of camera equipment,has received widespread attention and research.The digital image source camera device identification technology based on PRNU noise has made some progress,but its performance in practical applications is still limited by the quality of PRNU noise extraction.Therefore,improving the quality of PRNU noise is of great practical significance for achieving higher detection accuracy in digital image source camera device detection technology.The main research content of this thesis is to achieve higher detection accuracy of digital image source camera detection technology by improving the quality of PRNU noise.Firstly,this article summarizes and explains the novel and effective algorithms currently available.Secondly,the research approach of this article starts with the matching process of the digital image source camera device detection technology based on PRNU noise.Two different algorithmic approaches are proposed: one is to improve the quality of the camera device’s PRNU noise,and the other is to improve the quality of the PRNU noise of the digital image to be traced back to its source.This article conducted the following research in the study of improving the quality of PRNU noise:(1)This article proposed an algorithm based on block matching sampling to improve the quality of PRNU noise of camera equipment by removing correlation.Firstly,the interference noise in PRNU noise is estimated.This algorithm uses Principal Components Analysis(PCA)technique to make the reference PRNU noise and interference noise,which are in an ideal state,become uncorrelated and separable components.Before using PCA technique for training,a block matching sampling strategy is proposed to select samples with more image details to participate in PCA training.In the PCA domain,an appropriate shrinkage function is used to further estimate image details,which are then inverse transformed to the spatial domain.Finally,the camera equipment PRNU noise with higher quality is obtained by processing the estimated image detail residuals.The experimental results show that compared with other algorithms for improving the quality of PRNU noise,the algorithm proposed in this article has a higher accuracy in detecting the source camera of digital images.(2)The algorithm proposed in this article for improving the quality of PRNU noise in traceable source digital images is based on a dual residual convolutional neural network model with self-attention mechanism.First,the non-local self-attention mechanism is used to extract shallow features and obtain a larger receptive field.Then,the dual residual convolutional neural network structure is used to extract deep features,and two variable-sized convolution kernels are added to the network structure.Finally,the convolution kernels are used to reconstruct the PRNU noise features.This algorithm effectively combines the advantages of self-attention mechanisms and dual residual convolutional neural networks,enabling learning of PRNU noise features from different scales.The experimental results show that the proposed algorithm in this thesis outperforms other similar algorithms,and can extract higher quality PRNU noise from a single digital image.