Study On Reversible Data Hiding And Palette Images Based Steganographic Technologies

There has no doubt that the ever-changing multimedia technologies and products have brought great changes to human life. However, while we are experiencing the many advantages of digital terminal products, new challenges and opportunities have also attracted increasing concerns, in which information security issues are always the most important yet challenging topics. The conventional cryptographic technique leaves clear marks on the ciphertext for an eavesdropper to trace down, which may further lead the intentional attacker to interrupt this secure covert communication channel though he/she cannot access the original plaintext. As another security means, information hiding provides the ability to even conceal the presence of the on-going covert communication, while conveying a message to the intended receiver.Besides, with the rapid development of high-speed network and multimedia technologies, information hiding has become an important part of security technologies.Information hiding has its own place in the field of modern information security. It is not intended to replace the cryptography, but rather to supplement it. And, the ease of use,reproduction, edit and distribution of digital commercial products has promoted information hiding to become one of the most popular security research fields.Information hiding is actually an emerging and interdisciplinary research area covering different applications, among which watermarking and steganography are two very common use nowadays. In this thesis, we focus on some studies on steganography and reversible watermarking, where the primary contributions are:1. For steganography, we propose to use graph structure to model the steganographic operation for palette-based images, where palette colors are treated as vertices and edges between vertices represent a replacement relationship between the connected vertices. By applying depth-first search or breadth-first search, vertices corresponding to the palette colors are mapped to such different bit-strings that,any bit-string set that consists of some vertex itself and its neighboring vertices constitute a decodable codeword on the receiver side. Thus, extra information can be successfully embedded into the given palette image by modifying the actual colors of embeddable pixels. Experiments have indicated that, comparing with related works, the proposed method provides a larger embedding capacity or a lower embedding distortion while keeping a low computational cost.2. For reversible data hiding, in this thesis, we propose to adopt the prediction-error of prediction error of a pixel to reversibly hide the secret data into grayscale images. Specifically, after we compute out the prediction error of the pixel to be embedded, we further determine its prediction according to the prediction errors of neighbors. Thus, the second-order prediction error histogram can be used for data hiding. Experiments have shown that the proposed method can provide a relatively better performance for complex images. It is straightforward to extend the introduced technique to different media files, and/or the higher-order prediction errors.3. We propose a novel dynamic content selection-and-prediction framework to reversible data hiding, in which any illegal decoder will hardly recover the entire marked pixels and determine the marked prediction errors, providing a high security level when compared to the related works. Experimental results have also shown that the utilized data hiding operation can benefit from the proposed framework, and therefore provides a better payload-distortion performance.Actually, the proposed framework can be used/extended to the design of a reversible data hiding scheme to improve the payload-distortion performance. And,our framework can be also extended to steganography.4. We propose a reversible data hiding algorithm for encrypted palette images. The proposed scheme ensures that, the data extraction and image recovery process are separable and reversible, which is desirable for practical use. Our experimental results have shown that our scheme can provide a better distortion performance at the relatively low embedding rates, when compared to the related works.5. Neural networks can be regarded as the fitters of functions, implying that, in theory,we could use neural networks for steganography since the latter can be considered as a function of the cover, key and message bits. This thesis briefly discusses the possibility of neural networks used in steganography, and proposes to use neural networks to fit the simple LSB replacement and matrix coding steganography. We propose the idea of using two neural networks, i.e., policy network and embedding network, to jointly work for steganography.