DOUBLE JPEG Detection Using High Order Statistic Features

With the development of Internet and multimedia technology,there are a large amount of digital images in the Internet.Almost everyone can manipulate the digital images with the easy-to-use image processing software.Since images can be used in legal system,journalism and medical applications,image tampering can deliver misleading or even dangerous information for our society.This makes the digital image forensic technology more necessary.JPEG is widely used in various kinds of cameras and image processing software,if an image is JPEG format originally and it is saved as JPEG again after tampered,then it is JPEG compressed twice in the end.So some DOUBLE JPEG related works have been reported recently.This thesis focus on the DOUBLE JPEG detection using high order statistic features.The main research results are summarized as follows:(1)A DOUBLE JPEG detection method based on texture analysis and high order Markov transition probability matrix is proposed.Firstly,the image is divided into texture region and smooth region according to its content.Then,the difference arrays of the texture regions are calculated in order to enhance DOUBLE JPEG compression artifacts.The high order Markov transition probability matrices are used to model the difference arrays.After a thresholding technique,the elements in the matrices are fed into the classifier.Experiments results have shown that this method is effective.(2)A DOUBLE JPEG detection method based on DCT residual is proposed.Firstly,the decompressed image is convolved with 64 DCT basis,forming DCT residuals.Then the DCT residual features are built by quantizing the absolute values of all elements in the DCT residuals and collecting the histogram for each mode and each relative position.This approach can be interpreted as a projection model in the JPEG domain and can detect the DOUBLE JPEG effectively,especially in the situation where Q1 is very high.(3)A DOUBLE JPEG detection method based on Hadamard residual is proposed.In this method,the Hadamard basis is taken place of the DCT basis,and we use different quantization steps for different relative positions.The accuracy is as high as the accuracy of DCT residual based method,but it has low computing complexity and takes up less space.