Research On Steganography Of Three-Dimensional Model And Steganalysis Of Digital Image

With the rapid development of3D applications,3D model has been a new potential kind of steganographic media like text, image, video, audio. However, with the feature of small data capacity, irregular sampling and disorder of the vertex, steganograpy of3D model faces great challenges of improving the embedding capacity and obtaining low distortion, which needs to be further studied. Steganalysis is the art of attacking against steganography aiming at detecting the existence of hidden information, estimating the length of hidden information and even extracting hidden information. For its importance in the areas of military, intelligence and national security, steganalysis has been taken into account by information security researchers around the world.Therefore, the presented dissertation, which is supported in part by the National High Technology Research and Development Program of China (Grant no.2009AA012420) and Natural Science Fund of Guangdong Province (Grant no.9151009001000059), mainly focuses on the research on steganography of3D model and stegananalysis of digital image. The deep study on geometric features and representation of3D mesh model and3D point cloud model is made and three kinds of steganography for3D model are proposed with the purposes of designing the steganographic method whose invisibility and embedding capacity can be a balance point, the targeted research on steganography of color image and steganography in transform domain is made in order to propose effective universal blind steganalytic algorithm and special steganalytic algorithm for color image steganography and steganography in transform domain. The main research work in this dissertation is as follows:1. Two kinds of novel simple adaptive steganography for3D mesh model based on frame transform in spatial domain and transform domain are proposed. On the basis of the frame sampling of the video streaming, the presented algorithms use the frame-based transform space subdivision to map the3D spatial data to2D frame data. In the spatial embedding process, vertex weight estimation is used to estimate the degree of smoothness or roughness properties of3D model with respect to human visual system and adaptive embedding estimation to remove restrictions of fixed embedding size in each vertex and adaptively determine the vertex capacity, then multi-QIM is designed to embedding message. In the transforming embedding process, the just noticeable distortion (JND) analysis in wavelet domain based on the human visual system (HVS) is expanded to adaptively determine embedding strength, then generalized gray image is constructed and the Hidden Markov Model (HMM) in wavelet domain is applied to embed data in the zero-tree structure of HMM. The experiment results have demonstrated that the proposed technique are simple, adaptive, secure, have high capacity and low distortion.2. For3D point cloud model steganography, on the basis of deep study on the self-similarity compression of the3D point cloud model, a novel high-capacity steganography of3D point cloud model based on self-similarity compression is proposed. Based on the geometric feature of similar pattern or structure on the surface of3D point cloud model, the algorithm partitions the3D point cloud model to patches, clusters patches into similarity patch chains using self-similarity measures, and generates the codebook. The representative patches and similar patches are then taken from the codebook for every similar patch chain as the reference patches and the message patches. Finally, every message point in the similar message patches which has the point-to-point correspondence with a certain reference point in the reference patch can embed at least four bits using the proposed self-similarity position matching procedure (SSPM) by shifting the message point from its current point to the corresponding embedding position, which is computed over virtual sphere with the reference point as the centre. Experimental results show that the proposed algorithm is secure, has high capacity and low distortion which can embed at least four bits to embedding vertex.3. A universal steganalysis for color image steganography is put forward. Through the theoretical analysis and experiments, the embedding message affects the correlation among different color channels of the color image is giving. Therefore, the key of the algorithm is the Hilbert-Huang-Transform based analysis of color gradient orientation sequences between any two color channels and color gradient sum sequences of the color image. The Hilbert spectrum based characteristic vectors are constructed via empirical mode decomposition (EMD) of two categories of color gradient sequences and the SVM classifier is used to classify. The experimental results have demonstrated that the proposed method realizes the reliable steganalysis of color images with higher correct rate, shorter vectors and lower false positive rate compared with traditional color image steganalysis algorithms.4. On the basis of research on the vibrations of the DCT coeffience histogram of JPEG images, a steganalytic method is proposed to attack JPEG MB2steganography. The proposed method uses DCT coefficient histogram differential sequences as a measure of the correlation between JPEG images and its Cauchy fitting distributed model, applies Hilbert-Huang-transform based sequence analysis and constructs SVM classifier to discriminate the stego from the cover. The experimental results have demonstrated effectiveness of the proposed steganalytic method in attacking JPEG MB2steganography.5. A universal steganalysis is proposed to attack steganography in wavelet domain. The key of the algorithm is to model the wavelet coefficient of the image with two dimensional hidden Markov Model (HMM) model, calculate the HMM parameter sets based on hidden Markov tree (HMT) structure which descriptes the state correlation among the wavelet coefficient. The HMM parameter sets based characteristic vectors are used to construct SVM classifier. The experimental results have demonstrated effectiveness of the proposed steganalytic method in attack steganography in wavelet domain such as QIM, MFP and BPCS steganography in wavelet domain.