Application Of Compressed Sensing In Image Steganalysis

Currently,image space domain steganalysis is the latest technology for information hiding communication,and image space domain feature extraction is another important step in steganalysis.This paper mainly studies the image space domain feature extraction and universal steganalysis algorithm based on compressed sensing theory,the main results of the research are as follows:1.The structure of compressed sensing overcomplete dictionary has important influence on the perceptual performance of image spatial domain features,so,the design of overcomplete dictionary is one of the key issues in feature extraction of image spatial domain.This paper combines the sparse representation of block compressive sensing theory and its overcomplete dictionary.An adaptive fingerprint image airspace feature implementation method is proposed.This method is based on the random projection energy distribution of fingerprint image blocks,structurally adaptive multi-component sparse representation model for images,and classify it as a homogenous region with smooth,edged,and textured structures,a multi-component overcomplete dictionary consistent with its structural form was designed.When the image is reconstructed,utilizing the prior knowledge of the multi-component structure and sparse representation under the dictionary.Experimental results verify the effectiveness of the method.2.Based on the textural features of grayscale images,a scheme of the universal steganalysis is proposed using compressive sensing(CS)technology in spatial domain.In this paper,directional lifting wavelet transform is applied as a sparse representation,and corresponding sparse coefficient as statistical histograms for images.Then,measurement matrix of the CS is designed by co-occurrence matrix based generalized gaussian distribution model,where the measurement matrix will be utilized to sense sparse coefficients,and CS measurement values are regarded as textural features of the images.Finally,textural features are extracted for steganalysis implemented by the support vector machine.Extensive experiments are performed on four diverse image databases and by seven typical spatial domain steganographic algorithms.The results reveal that the proposed method is universal for detecting spatial domain steganography.By comparison with other well-known steganalysis algorithms,the method can provide better performance under most circumstances,and reduce features' dimensions.