Research On Analysis And Detection Of Steganography Techniques

This dissertation is mainly focused on analysis and detection techniques of steganography in images. The concepts in this dissertation keep abreast of the international steganography research progress. On the basis of probing into the essence of steganography systems and analyzing the existing steganography software and the latest information in the world, this dissertation proposes several practical and reliable algorithms. The main contribution of this dissertation are summarized as follows:Firstly, as the main research field of steganography usually takes the image as its carrier, on the basis of classifying and analyzing of the image carrier, this dissertation focuses on some basic statistics characteristics. These characteristics in carrier images include the random distribution of each bit plane, the correlation of bit planes, the difference charateristic in natural images, and also the coefficient distrubition of image transform domain .Five new detection algorithms which are aimed at LSB Stagonoraphy Algorithm have been proposed. These new algorithms, which have good detection performance of stabilization and correct ratio and can be used in grayscale image and color image detection, are based on the detection of characteristics of carriers (usually image carrier). The carrier characteristics include the correlation between bitplanes, the energy relation between pixels and the correlation between adjoint pixels. These five new spatial algorithms include:(1)Fliping Detection Based on Laplace Filter, which establishs a relation between LSB Flipping Detection and Image Laplace Filter, so as to get high accuracy and reliability in detection of Sequential LSB Replacing Stagonoraphy Algorithm.(2)the Gradient Energy - Flipping Ratio Detection. In this algorithm, the gradient and gradient energy of the image are defined and the flipping algorithm to detect trial is established. It realizes a high accracy and reliability in the detection of Sequential and Random LSB Image Replacing Stagonography Algorithm. It has the advantage of simple and easy implementation. Compared with RS Algorithm, it has a similar accuracy and a more speedy detection.(3)Flipping Ratio Detection Based on Extended Gradient Energy. In this algorithm, the gradient of image and the definition of gradient Energy is extended. This algorithm has a more stable detection performance.(4) Flipping Ratio Detection Based on Self-Relation. The self-relation of image pixels and flipping are used in this algorithm.(5)Pattern Recognition Detection based on Multi-Characteristic. In this algorithm, multi statistics of the image are extracted and the stego-images are classified by minimum square error linear discriminator.These algorithms contribute to the analysis and detection in the following aspects: make sure whether there is any hiding information in the target carrier or not; identify the steganography algorithm used in the carrier; estimate the length of embedded information; estimate the position of the embedded information, and so on. These algorithms are mainly designed for Image LSB Replacing SteganogrophyAlgorithm, but also can be used in a group of information hiding algorithms and schemes. As shown in the experimental results, these algorithms have good efficiency, applicability and accuracy.Then, this dissertation presents two new detection methods in transform domain. These two algorithms are aimed at Transform Domain Steganogrophy Algotithm in image steganogrophy. The basis of these two methods is high-order discrete wavelet decomposition of images in multi-resolution and multi-dimension. These two detection methods are:(6)the Higher Order Coefficient Detection based on Fisher Linear Classification;(7)and the Higher Order Coefficient Detection based on Support Vector Machine. The advantage and disadvantage of different kinds of detection algorithms are analyzed and compared in this dissertation.At last, the carriers are classified and analyzed from the view of Stegaganalysis. This dissertation establishes a theoretical model of the hu...