【作者】 曾宪庭；

【导师】 平玲娣； 潘雪增；

【作者基本信息】 浙江大学 ， 计算机科学与技术， 2010， 博士

【摘要】 随着信息技术的发展,信息的存储、加工、复制、传送变得越来越方便快捷。然而,在开放的网络环境中传输的各种信息很容易被窃听或遭到拦截；所有数字化信息很容易被复制、删除、剪切、拼接、篡改甚至伪造。因此,保障信息的安全就成为了人们不得不面对的一个巨大挑战。作为一种基本的信息安全工具,信息隐藏技术在信息安全领域中发挥着重要作用。然而,通常意义下的信息隐藏技术,由于其在载体中嵌入了机密信息,会导致原始载体永久性的失真,这些技术在数字媒体内容保护领域没有优势。因为如果嵌入一些信息使得这些数字媒体永久失真的话,就会导致这些媒体失去原有价值。因此,如何能够在原始媒体没有明显降质的情况下嵌入机密信息,并且当提取出机密信息后,原始媒体能够完全复原就显得十分必要。满足这种特性的技术称为无损信息隐藏技术。另外,如果这些能够复原的隐秘图像在遭到一些并无恶意的改变(比如：图像有损压缩或在噪声环境中无意添加了噪声)之后,仍能够精确地提取出嵌入的信息,将进一步扩大这种技术的应用范围,使这种技术能够从无损的应用环境扩展到有损的应用环境中去。满足这种特性的技术称为鲁棒无损信息隐藏技术。本文以图像为主要研究对象,重点研究了当前信息隐藏领域的新热点：大容量无损信息隐藏技术和鲁棒无损信息隐藏技术。本文主要取得以下研究成果。(1)提出了一种基于块固定参照像素的大容量无损信息隐藏方案。该方案能实现高嵌入和低失真。另外,该方案简单易行,除了机密信息长度之外,无须任何额外信息就可以从隐秘图像中提取出机密信息同时无失真地恢复出原始载体图像。实验结果表明,在所有隐秘图像没有明显降质(PSNR均超过30 dB)的前提下,8幅常用的标准灰度图像的平均纯嵌入容量超过0.97bpp。(2)提出了一种基于扫描路径的大容量无损信息隐藏方案。通过定义九种基本的扫描路径并计算各路径的相邻像素差,可以获得全方位的相邻像素差,选择最优路径的差直方图移位可以产生大量的冗余空间,因此能实现大容量信息的无损嵌入。实验结果表明,在8幅常用的标准灰度图像的测试中,在所有隐秘图像的PSNR均超过30 dB的前提下,平均纯嵌入容量超过1.0 bpp。(3)在块固定参照像素方案的基础上提出了一种基于块动态参照像素的无损信息隐藏方案。改进的方案利用块内排序后的中间像素作为参照像素,克服了块固定参照像素方案的不足。在保持了原方案的优点外,新方案大大地提高了嵌入容量。实验结果表明,块动态参照像素方案明显好于块固定参照像素方案。在含有1338幅灰度图像的UCID图像库(第2版)的测试中,在平均PSNR达到30.8 dB的前提下,块动态参照像素方案的平均嵌入容量超过了1.1 bpp。(4)提出了一种基于统计量移位的鲁棒无损信息隐藏方案。该方案通过引入一个统计量,并且使这个统计量产生一定的位移来嵌入信息。由于位移量及移位规则固定,所以这种方案是可逆的。另一方面,由于信息1和信息0分隔在不同的区以及统计量固有的特性,由有损压缩等非恶意攻击所产生的某种改变不会导致嵌入信息发生变化,因此,这种方案又是鲁棒的。实验结果表明,该方案与传统的方案相比有明显的优势。(5)提出了一种改进的基于统计量移位的鲁棒无损信息隐藏方案。原方案的嵌入机制仅允许块内一组像素发生改变,这会导致两个不利的影响。其一,要将统计量移位到不同的区以实现信息嵌入,往往要将其中的一组像素改变很大,影响了图像的质量。其二,对隐秘图像进行有损压缩时,机密信息抗压缩的能力降低,影响了算法的鲁棒性。改进的方案采用新的嵌入机制,让块内两组像素均发生改变。实验结果表明,与原方案相比,无论图像质量还是鲁棒性,改进的方案均有明显的提高。更多还原

【Abstract】 With the continuing advancements in information technology, information storage, processing, reproduction, transmission becomes more and more convenient. However, all kinds of information in an open network environment can be easily eavesdropped and intercepted; all digital information can easily be copied, deleted, cut, spliced, tampered or forged. Therefore, information security is faced with enormous challenges that we have to face. As a basic information security tool, information/data hiding plays an important role in many critical areas of information security.However, in most cases of data hiding, the cover media may suffer some permanent distortions as a result of data hiding and can not be inverted back to the original one. These techniques have no advantages in the field of digital media content protection because the images processed by these techniques will lose their values. Therefore, in some applications, such as media content protection, it is desirable to reverse the stego-image back to the original cover image without any distortion after data extraction, i.e., the technique not only can embed secret data into a cover image, but also the original cover image can be recovered from the stego-image with no distortion after the hidden data have been extracted. Techniques with this property are referred to as lossless data hiding. Furthermore, the robustness of a lossless data hiding algorithm against non-malicious changes such as lossy compression or sometimes unavoidable addition of random noise which is below a certain level and does not change the content of an image can be useful in the context of reversible data hiding. It might enlarge the application scope of lossless data hiding technique as it enables the lossless data hiding to convey information in a lossy environment. Techniques with this property are referred to as robust lossless data hiding or lossless robust data hiding. According to the problems mentioned above, this dissertation, after a deep study on related works, mainly focused on images and launched the research in two aspects: high embedding capacity lossless data hiding and robust lossless data hiding. The main contributions are as follows.(1) A reversible data hiding scheme based on fixed reference pixel of block is proposed. The proposed scheme can offer high embedding capacity and keep low distortion. Furthermore, the proposed scheme can be implemented easily and no extra information except message length is needed to extract data and restore image. Experimental results show that the average pure payload among eight commonly used images can be greater than 0.97 bits per pixel (bpp) while maintaining all the PSNR values of the stego-images above 30 dB.(2) A lossless data hiding scheme based on scan path is proposed. Nine basic scan paths are defined, and this means all-directional adjacent pixel differences can be obtained. The proposed scheme can offer high embedding capacity and keep low distortion by choosing the best scan path and the optimized pixel difference. As experimental results have shown, the average pure payload among eight commonly used images can be greater than 1.0 bits per pixel (bpp) while all the PSNR values of the stego images remain larger than 30 dB.(3) An improved reversible data hiding scheme using dynamic reference pixel of block instead of the fixed reference pixel of block is proposed. The improved scheme uses the reference pixel whose value is the median value of the whole sorting sequence of block. The improved scheme can offer higher embedding capacity while mantaning the advantages of the original version. Experimental results have shown that the performance of the dynamic reference pixel of block outperforms that of the fixed reference pixel of block. The improved scheme has been applied to all of the 1338 images in UCID-Image Database (v2) which includes many different types of images, and the experimental results show that the average payload among these images is larger than 1.1 bits per pixel (bpp) while keeping the average PSNR values 30.8 dB.(4) A robust lossless data hiding scheme based on statistic shifting is proposed, in which a statistic is introduced and bits can be embedded by shifting the statistic. The shift quantity and shifting rule are fixed for all blocks, and reversibility is achieved. On the other hand, owing to the separation of bit-0-zone and bit-1-zone as well as the particularity of the statistic, minor alteration applying to the stego-image generated by non-malicious attacks such as JPEG compression will not cause the bit-0-zone and the bit-1-zone to overlap, and robustness is achieved. Experimental results show that, compared with previous works, the performance of the proposed scheme is significantly improved.(5) An improved robust lossless data hiding scheme based on statistic shifting is proposed. In the original version, the embedding mechanism lets only one group of pixels of block change, leading to two disadvantages. First, to shift a statistic to different zones cause a group of pixels to be altered greatly, leading to a low image quality. Second, the hidden data can not resist the JPEG compression at high compression levels during lossy compression, leading to a weak robustness. The improved scheme uses a new embedding mechanism to block and two groups of pixels of block can be changed. Experimental results show that, compared with the previous version, the performance of new scheme is significantly improved in terms of image quality and robustness.更多还原