【作者】 欧博；

【导师】 赵耀；

【作者基本信息】 北京交通大学 ， 人机交互， 2014， 博士

【摘要】 随着因特网和多媒体信息技术的发展,存储、修改和传播数字多媒体信息(文本、音频、图像、视频等)变得越来越容易。人们可以很容易地下载这些多媒体信息,并加以修改和发布。这不仅损害了版权所有者的合法权益,也会引发关乎国家安全、企业机密和个人隐私等一系列信息安全问题。信息隐藏是一种隐蔽通信、版权保护和内容完整性认证的有效技术手段。近年来提出的可逆信息隐藏技术能够将特定信息嵌入到载体,并允许合法用户在提取该信息后无失真地恢复出原始载体,因而在军事、医学、司法等对图像内容敏感的领域受到了广泛关注。本论文以高保真的可逆信息隐藏技术为主要研究内容,探讨在给定嵌入容量的情况下如何更有效地减少嵌入失真,提升图像视觉质量。论文取得的主要创新性研究成果包括：1.提出了一种基于可自适应选择预测误差直方图的高容量可逆信息隐藏方法。该方法设计了两种预测器,通过参数调节得到不同的预测误差直方图。对比上层嵌入,自适应地得到当前嵌入层的最优预测误差直方图,使得像素补偿的收益最大化。实验结果表明该方法在高嵌入率情况下的图像质量明显优于基于直方图移位的传统方法。2.提出了一种基于偏微分方程预测器的可逆信息隐藏方法。偏微分预测器能够根据图像的局部相关性迭代更新得到最优预测值。在每一次迭代中,通过计算当前预测值在上下左右四个方向上的梯度,来自适应地分配权重,因而预测更准确。实验结果表明在同等嵌入率下该方法带来的失真更小。3.提出了一种基于非局部均值预测的可逆信息隐藏方法。通过全局地利用图像的自相似性来提升像素预测的准确性,不仅能在图像平滑区准确预测,也能在纹理区取得较好效果。实验结果表明,该方法对于自然图像尤其是纹理图像可取得更好的嵌入性能。4.提出了一种基于像素值排序的可逆信息隐藏方法。以像素块为基本嵌入单元,利用块内像素值的大小顺序在信息嵌入前后保持不变这一特性,将图像的嵌入划分了多个层级。通过容量划分和像素选择技术,根据容量自适应地在不同类型的像素块嵌入信息,得到最优的嵌入效果。实验结果表明该方法在小嵌入容量下对图像的改动更小。5.提出了一种基于二维直方图的可逆信息隐藏框架。利用预测误差间的相关性,将邻近的两个预测误差作为嵌入单元来设计一种二维的可逆嵌入方案,并给出了多个新的、更有效的二维可逆映射,理论证明了该方法的优越性。实验结果表明该方法明显优于当前主流的可逆信息隐藏方法。更多还原

【Abstract】 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.更多还原