Research On Image Tampering Detection Based On PRNU And Linear Pattern

With the rapid development of information technology,the functions of digital products such as digital cameras and mobile phones have become more and more complete,and people's living standards have been significantly improved.However,due to the emergence of image editing software,anyone can tamper with or falsify images according to their own needs,which causes serious social problems.Therefore,ensuring the authenticity and originality of digital images has become a concern of a wide range of scholars.For tampering forensics of digital images,experts and scholars at home and abroad have proposed a large number of algorithms.Among them,image tampering detection technology based on pattern noise has made great progress.The existing image tampering forensics technology has universality,mainly relying on camera inherent noise,namely sensor pattern noise(SPN),to identify tampering and locate tampering regions.However,the existing image tampering forensics technology based on pattern noise pays more attention to image copy-move,splicing,etc.,and relatively few image blurring and enhancement operations are detected.Based on the image tampering detection of pattern noise,this paper is related to the actual situation,two effective methods for image tampering forensics detection are proposed.The specific research results are as follows:1)The accuracy of the existing linear filter kernel estimation method is not high,a linear gaussian filter kernel estimation algorithm based on image PRNU(Photo-Response NonUniformity)noise is proposed.The method uses the camera reference PRNU noise as the identification fingerprint,extracts the image noise residual from the clean image and the test image,respectively,and calculates the cross-correlation between the two image noise residuals and the identification fingerprint.Mathematically,the linear correlation between the two crosscorrelations can be determined.Then,the filter kernel coefficient of the test image is estimated according to the linear least squares method,and the mean squared error between the filter kernel estimated by the linear least squares and the actual filter kernel will be calculated to measure the effectiveness of the algorithm.In addition,due to the unknown of clean images,how to obtain images without other additive noise is critical.The scheme proposes a method of recapture the test image,and the recaptured image is used as a clean image,which effectively improves the detection performance and is robust to JPEG compression.2)Aiming at the problem that the existing global contrast enhancement detection has low classification accuracy under the middle/low quality JPEG compression,an image contrast enhancement detection algorithm based on linear pattern is proposed.This paper proposes to extract the image noise residual from the image,adopt the blocking strategy for the noise residual,calculate the linear pattern of each block,and calculate the corresponding power spectral density according to the linear pattern of the corresponding image block to present the characteristics of the linear pattern.The mean power spectral density of the entire image is then calculated as a classification feature.Finally,the support vector machine is used for classification.The experimental results show that this scheme can be effectively classified and can resist a certain degree of image compression.In addition,it can effectively solve the local gaussian blur and contrast enhancement tampering location detection problem.