Research On Blind Forensics Method Of Local Deformation In Image Retouching

With the rapid development of computer technology and network,as well as the popularity of mobile phones,the convenience of images has made it widely recognized and disseminated as a carrier of information.However,using image processing technology to do something illegal will infringe upon citizens' right to know and the right to portrait.The local deformation of the image can be used to work on the slimming and thin face of the characters in the image.Long-term seeing the over-modified image will make people mistakenly think that the modified image is the normal image.The widespread spread of this retouched image will affect people's consumption.Views and values will even affect social stability to a certain extent.The local deformation of the image is different from the scaling rotation operation of the image,which does not introduce periodic changes in the pixel,but this deformation also causes a change in the pixel value.We studies the method of using high-order linear residuals and GLCM methods to detect local deformation of images.We extract the trilinear residuals from the tamper image,and then pass the quantization and truncation processing,then use the sliding blocks of different sizes to extract the GLCM features,and send the obtained features into the fuzzy c-means clustering(FCM)method to obtain the tampering of different blocks.The probability map is used,and the averaging method is used to fuse the tamper probability map.The fused operation is used to remove the inconsistent region.The experiment shows that the local deformation of the image can be detected by this method.We propose a multi-strategy fusion method based on first-order difference image and its texture features to detect the local deformation of the image.Firstly,the texture features of overlapping blocks of different sizes are extracted on a color channel,and input into the fuzzy c-means clustering method to generate tamper probability maps(TPM).The first layer of fusion combines multiple TPMs.Secondly,TPMs of different color channels and different texture features are respectively combined in the second layer and the third layer.Experimental results show that the method can accurately detect the position of local deformation of the image.The above two methods can detect the local deformation of the image,but the two methods could not obtain good detection results for the local deformation of the compressed image.Therefore,we propose a block-based method for detecting local deformation of compressed images.We subtract the tamper image from the falsified image to obtain the residual map,extract the block effect mesh from the residual map,and detect the local deformation of the image by calculating the mesh offset.This method obtains better detection results than using block effect detection directly for tampering images.