Research On High-fidelity Reversible Data Hiding

As the development of Internet and multimedia technology, it is possible for world-wide people to store, copy, edit and distribute multimedia data (including text, audio, image and video) nowadays. Peoples can freely download the digital content from the Internet, and then edit it for their personal uses. However, it harms the rights of copyright owners, and also brings a great number of information security problems. Data hiding offers a way of content protection. Recent years, reversible data hiding (RDH) is pro-posed to embed data imperceptibly into cover media in a reversible way, such that the authorized users can losslessly recover the original content after extracting the hidden message. It has aroused great concern in the medical, military, judical fields where any permanent distortion on the original content is strictly forbidden.In this paper, we mainly focus on the high-fidelity RDH which introduces less dis-tortion for a given capacity. The research achievements are listed as follows.1. Propose RDH using optional prediction-error histogram modification. By consid-ering the pixel compensation during the multiple layer embedding, an optional predictor is designed to generate the most appropriate prediction-error histogram, which results in less distortion at the same embedding rate. Unlike other histogram based schemes, the generated prediction-error histogram can be tuned through the selection of threshold for each layer to strike the balance between capacity and pixel compensation. Experimental results demonstrate that the proposed method introduces less distortion at high embedding rate.2. Propose RDH using partial differential equation (PDE) predictor. The general idea of PDE is to implement anisotropic diffusion by encouraging intra-region smooth-ing in preference to inter-region smoothing. As for the prediction in PEE, such property can also be utilized to make context pixels with high correlations being weighted larger than the ones with low correlations. Since PDE predictor can bet-ter exploit image redundancy, the proposed method introduces less distortion for embedding the same payload.3. Propose RDH using non-local means (NLM) predictor. By globally utilizing the potential self-similarity contained in image itself, the proposed method aims to achieve better prediction even in texture regions. The incorporation of NLM makes the proposed method possible to achieve accurate prediction in both smooth and texture regions. Compared with other methods, the proposed method can yield a better capacity-distortion performance, especially for texture images.4. Propose RDH using invariant pixel-value-ordering and PEE. In the method, an im-age is divided into non-overlapped blocks and the pixel block is used as the basic u-nit for data embedding. For each block, the maximum-valued (or minimum-valued) pixels are first predicted and then modified together such that they are either un-changed or increased by1(or decreased by1) in value at the same time. Compared with the prior art, more blocks suitable for RDH are utilized and image redundan-cy is better exploited. Moreover, a mechanism of advisable payload partition and pixel-block-selection is adopted to optimize the embedding performance in terms of capacity-distortion behavior.5. Propose pairwise prediction-error expansion (PEE) for efficient RDH, which for-mulates a new paradigm of PEE in a higher dimensional space to exploit correla-tions among adjacent prediction-errors. Here, every two adjacent prediction-errors are considered jointly to generate a sequence consisting of prediction-error pairs. Then, based on the sequence and the resulting two-dimensional prediction-error histogram, a more efficient embedding strategy, namely, pairwise PEE, can be de-signed to achieve an improved performance. The superiority of our method is the-oretically proved, and then verified through extensive experiments.