English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Tools annual

Global: Zendy Free Plan

Global: Zendy Tools monthly

Global: Zendy Plus monthly

According to the image processing theory, complex texture image regions have larger entropy values, an adaptive steganography algorithm in the sparse domain is proposed here. The algorithm first divides an image into nonoverlapping image blocks, then computes the entropy of each block and chooses the mean of all entropies as a threshold. Image blocks whose entropy is higher than the threshold are selected for sparse decomposition as complex texture image regions. The secret message is embedded into the decomposition coefficients, and then the stego image is reconstructed with modified coefficients. The experimental results prove that, given the same payload, the proposed adaptive algorithm has a better performance against the steganalyzer. Introduction Steganography is a technology that tries to embed secret messages to general objects(images, audio, etc.) so as not to attract outside attention. As a main adversary, steganalysis aims to detect the presence of hidden messages. Existing steganalysis algorithms often rely on the statistical characteristics of image signals, but ignore the change on the semantic level 0. This inspired researchers to embed secret information in the semantic content of a carrier signal to improve the security. Image sparse representation is capable of representing the image in a sparse and efficient way, so it can be applied to image steganography. Currently only a few papers preliminarily explored the application of image sparse representation to image steganography. Cancelli embedded secret information by modifying the decomposition coefficients [2,3,4]. It is demonstrated that the same embedding algorithm applied in the sparse domain is more secure can it is in the pixel domain. However these papers just applied sparse representation to image steganography, did not take the influence of other factors, such as images content , into account. And studies have found that if better steganalysis tools are used, such as the SPAM (subtractive pixel adjacency matrix) steganalyzer0, the possibility of detection will increases greatly. Considering these problems, an image adaptive steganogaphy algorithm based on sparse representation and entropy is proposed. The rest of the paper is organized as follows. In Section II,a brief introduction of steganography using image sparse representation is given. The proposed algorithm is presented in Section III. Section IV reports the results of the experiments that we carried out to validate the proposed technique. Conclusions are given in Section V. Related Work In the sparse representation theory, each signal is represented with a linear combination of few basic elements. The elements N i R  g for the decomposition of signals are called atoms, the set of atoms   K i i g D ,.., 1   is known as a dictionary. If K>N, atoms are linear correlated, the dictionary D is redundant. There exists two critical issues to apply sparse representation to image steganography: the International Conference on Information Science and Computer Applications (ISCA 2013) © 2013. The authors Published by Atlantis Press 214 construction of the dictionary D for image sparse representation and how to get the efficient and unique decomposition coefficients of an image given the dictionary D. The following provides solutions for these two problems. Construction of The Redundant Dictionary. The goal is to embed secret messages at a semantic level, so the required redundant dictionary should be as semantically meaningful as possible. Generally, the dictionary can be constructed through the following three methods: 1) generated manually [4]; 2) generated directly using existing generating functions, like DCT transform, Haar transform [6], and Curvelet transform [7,8]; 3) generated by learning, the dictionary is learned given training samples. Here the K-SVD (K-means Singular Value Decomposition) 0 algorithm is used to obtain the redundant dictionary. The objective function can be expressed as: } { min 2 D, F X DX Y  s.t. , , i 0 0 i T x   (1) Where K N R D   is the dictionary, n i i y Y 1 } {   is the set of n training signals, n i i 1 } x { X   is the coefficient matrix for the sparse representation of signals. T0 is the maximum number of non-zero elements for sparse decomposition coefficients. The solution of equation (1) is an iterative process. First, we find the coefficient matrix X for signals Y assuming D is fixed. Then, the dictionary is updated according to the coefficients. The update of the dictionary is performed by column. Take the update of the kth atom and its corresponding coefficient k T x (the kth row of X) for example: 1)Find the corresponding indices of all the sample signals that use the atom k g in } {yi and form a set, defined as } 0 (i) x n, i 1 | {i w k T k     . 2)Compute the overall representation error matrix,Ek,by     k j j T j x g Y . Ek (2) 3)Define k  as a matrix of size k w N  ,with ones on the (w(i),i)th entries and zeros elsewhere. Use k k R E   k E to find error columns that correspond to samples that use the atom k g . 4)Apply SVD decomposition T R V U  k E .The updated dictionary column is ), 1 (:, gk U  (3) the coefficient vector k k T k x   R x is ) 1 , 1 ( ) 1 (:,  V x R (4) Repeat until convergence(stopping rule).Note that, the atoms of the learned dictionary is real-valued. This will result in a quantization error when the stego image is reconstructed,so the atoms should be further binarized with only the 0 and 1 values are allowed. Stable Sparse Decomposition Algorithm for Image Stegonagraphy. The commonly used MP(matching pursuit) algorithm [7,8] is a greedy algorithm that selects the dictionary atoms sequentially. It can be described as two steps: , ) , ( ] , [ 1 D R S g c k k k    (5) , ) , , ( 1 k g c R U R k k k    (6) where S(·) is a selection function and U(·) is an update function. k R is the residual signal, ck is the specific weight of the kth atom , k  is the index of the kth atom of the decomposition. However it's not suitable to directly apply the MP algorithm to image steganography, because the insertion of the message may cause the instability of decomposition path and decomposition coefficients. To solve the problem, Cancelli defined the selection rule as follows0. , } 0 : 0 { max c 1 k       k k k cg R R c  (7)

An Image Adaptive Steganogaphy Algorithm Based on Sparse Representation and Entropy