结合人类视觉特性的小波域图像数字水印算法研究

【作者】 陈康；

【导师】 方旺盛；

【作者基本信息】 江西理工大学 ， 通信与信息系统， 2010， 硕士

【摘要】 本文主要研究了基于小波变换的数字图像水印技术。系统地介绍了数字水印的研究背景及意义、研究现状、特征和分类,给出了图像数字水印技术的基本框架、常见攻击、评价标准以及典型算法,并简要的概述了小波分析理论。本文提出了几种结合人类视觉特性的小波域图像数字水印算法。提出了一种基于人类视觉系统的小波域图像水印算法。算法利用人类视觉特性和小波变换的特点对小波域JND门限模型进行改进,并计算载体图像各小波系数临界可见误差(JND)门限,通过选择一组JND门限较大的小波系数嵌入数字水印信息。实验结果显示该算法具有很强的鲁棒性和不可见性。由于各个小波系数的嵌入强度是一样的,算法较强的抗攻击性能是建立在水印透明性有一定程度破坏的基础上,如果水印过大这种破坏就比较明显。因此,该算法适应于小容量水印系统。提出了一种基于水印容量的小波域图像水印算法。确定水印嵌入模型,通过修改小波系数实现水印的嵌入,水印嵌入强度由JND门限计算得到。由此,利用水印容量公式建立该水印模型下载体图像水印容量的估计方法;并根据水印攻击类型和强度,对载体图像的水印容量进行估计值,根据此估计值调整水印的大小。针对水印攻击类型和强度无法确定情况下,通过对水印容量公式的分析,选择水印失真上限较高的一组小波系数嵌入水印信息提高水印鲁棒性。实验表明该水印算法具有较好的不可见性,对于常见的水印攻击如:添加噪声、图像滤波和JEPG压缩有很强鲁棒性。由于算法利用混沌系统预处理水印,使得算法具有较高的安全性并且对图像剪切攻击也有较强的鲁棒性。对于其他几何攻击下,算法提取的水印效果不理想。提出了一种基于小波变换的抗几何攻击图像水印算法。介绍了几种常见的抗几何攻击水印方案,本文选择基于图像特征抗几何攻击方案实现算法的抗几何攻击性能,将小波分解的低频子图作为载体图像的图像特征嵌入水印信息。通过对低频子图的小波系数的分块量化实现水印的嵌入,并且根据人类视觉特性和几何攻击特点增加低频子图纹理区小波系数的量化步长。水印提取过程:将水印图像的低频子图进行分块,根据多数原则判定各个子块所含的1bit水印信息。试验结果表明,该算法具有水印容量大,水印盲提取,水印不可见性好以及抗攻击性能好的特点,其中算法抗攻击性能包括算法对常见几何攻击和添加噪声、JEPG压缩等有较强的鲁棒性。更多还原

【Abstract】 This dissertation studied technique of image digital watermarking based on wavelet transform. The purpose of the research, the history of the watermarking technique, the development status of multimedia watermarking and the characteristics and classification of watermarking were reviewed. This dissertation surveys systematically, the basic frame of the watermarking system and attacks classification and typical algorithms. Then, this dissertation introduces the basic knowledge of wavelet transformation including the background of wavelet analyzing, consecutive wavelet transformation, discrete wavelet transformation, and multi-differentiating analyzing and Mallat algorithm. On this basis, several image digital watermarking algorithms based on wavelet transform were presented.This dissertation introduced a wavelet watermarking method based on human visual system (HVS). In the algorithm, using the HVS to compute the JND threshold in wavelet domain of original image, the watermark information were embedded into a set of wavelet coefficients of lager JND threshold. The experiments showed that this algorithm had strong robustness and invisibility. As the embedding intension of each wavelet coefficient was the same, the strong resistance to watermark attacks of this algorithm wad based on watermark transparency damaged, If the watermark was too large, such damage was more evident. Therefore, the algorithm was adapted to small-capacity watermarking system.This dissertation studied the image watermarking capacity analysis in wavelet domain. First, the watermark embedding model was determined, which was achieved by the rule of modifying the wavelet coefficients, the watermark embedding intension was calculated by JND model. Then, the estimation of watermarking capacity were established by the watermarking capacity formula in this watermark embedding model; By the watermark attack type and intensity, watermarking capacity estimates of original image could calculate. Algorithm could adjust to the size of watermark by the estimate. Generally, as the type and intensity of watermark attacks could not be determined, a group of wavelet coefficients with lager allowable distortion was chosen to embedded watermark based on the watermarking capacity analysis, the watermark embedding intension was decided by allowable distortion. Experimental results showed that this algorithm could get a watermarked image with perceptual invisibility. Moreover, the algorithm was robust to typical watermark attacks such as: noise adding, image filter and JPEG compression. As the algorithm used chaotic system to pre-process watermark, algorithm had high security and robustness to cropping attack. Whereas, the effect of watermark extraction by this algorithm under others geometric attacks were not perfect.A geometric attack resistant image digital watermarking algorithm based on wavelet transform was proposed in this dissertation. First, several typical geometric attack resistant watermarking scheme were introduced. This algorithm regarded the low-frequency subband of original image as the image character to embed watermark. The low-frequency subband was divided into many blocks, and then 1 bit of watermark information was embedded into each block repeatedly by quantization, and according to the characteristics of human visual system, we increased the quantization step of wavelet coefficients in highly textured areas. For watermark extracting, the 1 bit of watermark information was obtained by majority rule from each block. The watermark capacity of the algorithm was large and we didn’t need the original image to extract watermark. Experimental results showed that the algorithm could effectively resist geometric attacks in various ways, and it also strongly robusted against the common attacks such as noise adding and JPEG compression etc.更多还原