Research On Blind Digital Image Forensics Methods Based On Non-separable Wavelet

With the advent of the era of network, people are access to all kinds of information more and more depend on network. And the proportion of digital images in network information also increased year by year. With the wide application of digital images, image processing technology also more and more easy to operate. In recent years, image tampering cases emerge constantly in the press, law, science and other fields. It has seriously affected the normal order of all walks of life and caused the public to question the authenticity of the internet information. Under the background of the above mentioned, blind digital image forensics that is a new research field appeared.It’s difficult to get a unified detection scheme for passive image identification technology. Because, image forgery technology has the diversity and complexity. At the same time, blind digital image forensics technology involving multiple disciplines. Image blind forensics has frontier and interdisciplinary, which makes the research work has the unprecedented challenge.The aim of the paper is to detect the authenticity of digital images. The following is completed work:1. First of all, the paper analyzed the deficiency of the traditional digital image identification technology. Then the paper detailed introduces the basic theoretical framework of blind forensics technology, the current main identification algorithm and problems. Some algorithm are put forward through analyzers the tamper with the technology in this paper by myself. At the same time, it is combined with the advantages of non-separable wavelet in the time-frequency analysis.2. To the type of copy move image forgery, a blind forensics algorithm that based on non-separable wavelet and Zernike moment is proposed. Non-separable wavelet is shift invariant and therefore more suitable than discrete wavelet transform(DWT) for data analysis. Firstly, the input image is decomposed into approximation(LL1) and detail(HH1) sub-bands. Then the approximation sub-band are divide into overlapping blocks and extracted Zernike moment feature. Using threshold, matched pairs of the image block using cosine similarity correlation coefficient. The results showed that the effectiveness of the proposed method better than the method using Daubechies wavelet.3. To the type of computer generated image, firstly, the paper introduced the forming process of natural images and computer generated images in the paper. Then the difference of the high frequency sub-bands histogram of the image that decomposed by two kinds of wavelet is compared. Finally, the paper proposed an image detection algorithm based on non-separable wavelet and descriptive statistics for computer generated images. The main idea of the algorithm is firstly decomposition the input image using non-separable wavelet of image. Then obtain the statistical feature from the detail component. Finally, using LIBSVM tool package training and classification of image data. This experiment adopts the standard image database provided by Columbia University. The results show that this algorithm has good classification accuracy.