Application Of Sparse Representation And Encoding In Steganography

Steganography, as an important information hiding technique, has received great attention and wide research. The purpose of steganography is to embed secret information in multimedia data to achieve covert communication.Digital images,as one of the most common multimedia data transimitted in the Internet, are much ideal to be used as carriers.Image steganography has been studied for many years.Some practical methods have been proposed during the period,but there still exist some unsolved problems. Here, with grayscale images as carriers,we respectively studied image steganography in sparse domain and spatial domain. The main contributions of our study are as follows.(1) Steganography with distortion function based on sparse representationSteganographic algorithms based on sparse representation usualy use decomposition coefficients to embed secret information. Modification of one decomposition coefficient usually will affect more than one pixel’s value. In order to solve the problem, we introduced a distortion function. We choose the coefficients modification direction to minimize the distortion function. Experiments show that, with the same embedding payload, the proposed method is more secure than previous steganographic algorithms in sparse domain.(2) Adaptive steganographic algorithm based on adjacent pixel value differenceHere we randomly organize adjacent pixel pairs in different directions as embedding units, compute the threshold parameter according to the embedding payload requirement, choose pixel pairs with difference values not less than the threshold and embed the secret information using EMD algorithm.The synchronization of the sender and the receiver is achieved by data readjustment. Experimental results show that, under different steganalysis methods, the anti-detection performance of this proposed method outperforms other adaptive steganography algorithms and encoding schemes.(3) Adaptive syndrome trellis code based on adjacent pixel value differenceHere we use the adjacent pixel value difference of the prediction-error image to measure the distortion profile of pixels in the cover image.The syndrome trellis code is used here, with the weight of every edge in the trellis graph set as the distortion profile of the corresponding pixel. The path with minimum weight is searched, so that the pixels can be used selectively. It can be shown in the experiment results that, this algorithm can further improve the security of steganography.