|
发布时间: 2019-09-16 |
图像处理和编码 |
|
|
收稿日期: 2018-12-21; 修回日期: 2019-04-22
基金项目: 国家自然科学基金项目(61462013,61661010)
第一作者简介:
陈怡, 1996年生, 女, 硕士研究生, 主要研究方向为医学图像处理、信息安全。E-mail:Yi-Chen.gzu@outlook.com;
张健, 男, 副教授, 主要研究方向为图像融合。E-mail:34044200@qq.com; 王国美, 女, 副教授, 主要研究方向为图像分割。E-mail:15885076515@163.com.
中图法分类号: TN911.73
文献标识码: A
文章编号: 1006-8961(2019)09-1434-16
|
摘要
目的 弥散加权成像(DWI)作为一种新型医学影像成像技术,已逐渐成为诊断心脏、大脑、肾脏、肝脏等器官中的神经、纤维组织病变的重要方法和手段。与传统的核磁共振(MRI)成像相比,通过使用不同的扩散方向矢量,在不同的扩散参数下,DWI图像呈现的灰度信息也有所不同。目前尚无相关文献提出针对DWI图像版权信息进行有效保护的相关研究。方法 为有效保护病人的DWI图像版权信息,提出一种基于DWI图像的整数小波变换域(IWT)统计直方图的鲁棒水印算法。该算法首先通过最大类间方差分割算法和面积控制阈值获取指定断层中带有弥散梯度方向图像的前景区域,作为待嵌入区域。对待嵌入区域使用整数小波变换获取低频子带系数,利用固定步长对低频子带系数进行统计,生成统计直方图,对统计直方图相邻簇的比值关系进行修改用于水印嵌入;最后提出DWI表观系数与弥散张量成像(DTI)中弥散张量值的可逆关系构建可逆密钥,利用该密钥将嵌入水印后的DWI图像再次加密,从而有效保护DWI图像的版权信息。结果 实验结果表明该算法引入的水印信息对DWI图像中的纤维参数改变量极小。在纤维方向和平均弥散程度改变个数上,本文算法与文献方法相比,分别降低了100多个和30多个;在可视质量上,本文算法提高约8 dB。在高斯噪声、小角度旋转等攻击中,本文算法能够提供较高的提取水印准确率。结论 本文算法对医生诊断的影响在可接受的范围内,且在感兴趣区域遭受各种常见攻击时,具有较高的安全性和鲁棒性。
关键词
弥散加权成像; 直方图; 固定步长统计; 比值关系; 鲁棒水印
Abstract
Objective Diffusion weight imaging (DWI), as a new medical imaging technique, transforms the diffusion motion of water molecules in tissues into grayscale or other parameter information of an image by applying multi-directional diffusion magnetic sensitive gradients under each diffusion sensitive gradient. This technique can be used for the auxiliary diagnosis of living heart myocardial fiber modeling, brain fiber, lesions of the central nervous system, liver fibrosis, and other diseases. With the popularization of telemedicine diagnostic technology, an increasing amount of DWI data are being used for remote diagnosis and scientific research. DWI images, which are originally stored and used on a single machine in a hospital, must be transmitted and used over the network. Many scholars have proposed many watermarking algorithms for protecting medical images, such as the reversible watermarking algorithm, robust reversible watermarking algorithm, and zero-watermarking algorithm. The advantage of the reversible algorithm is that it can be completely used for nondestructive image recovery. The robustness of the reversible watermarking algorithm is too weak to guarantee the existence of the reversible watermark when embedded images are attacked intentionally or unintentionally. Therefore, some researchers propose the robust reversible watermarking algorithm. The robust reversible watermarking algorithm could restore an original picture when no attack occurs and could draw embedded watermarking. It ensures that its robust reversible performance should carry additional information. Thus, it must consume a large amount of transmission bandwidth. Some robust reversible watermarks are constructed by dual watermarking, and they depend on one another's information to extract the watermarks. To protect medical images by other methods, some researchers use the zero-watermarking algorithm, which is different from the traditional method that embeds information into images. The zero-watermarking algorithm can retrieve internal features from data to build binary watermarking and then save it in a third-party application. When the image is used by other people without the license, we could use zero-watermarking to prove copyright. Thus, the zero-watermarking algorithm, as a non-embedded algorithm, cannot actively complete the protection of property information. The robust watermarking algorithm plays an irreplaceable role in ensuring that the medical image watermarking information has certain robustness. To prevent unauthorized DWI images from being used or tampered with, this study proposes a robust watermarking algorithm based on DWI images. Method The algorithm initially obtains specified slips by the maximum inter-class variance segmentation algorithm and area control threshold to ensure that the selected slice has a sufficient embedding area because the tip and the bottom of the heart are unsuitable for embedding. The foreground region with a diffusion gradient direction image is prepared for embedding. We obtain the low-frequency sub-band coefficient by using integer wavelet transform in the default region. Then, we count the low frequency and analyze the low sub-frequency coefficient by using the fixed step length; the low sub-frequency coefficient follows the characteristics of the coefficient of DWI images. The ratio relation of adjacent clusters in the histogram subject area is adjusted for watermark embedding. Finally, we propose to design the quantitative reversible relationship between apparent DWI coefficients, with diffusion tensor imaging (DTI) as the key. We use this key to encrypt a DWI image after embedding the watermark to effectively protect the copyright information of the DWI image. Result The algorithm can maintain its robustness and reduce the change in the DTI parameters in the experiment on robustness and changes in the parameters of DTI after embedding. The proposed algorithm also has excellent robustness in attack experiments, such as those involving Gaussian noise, contrast expansion, and small angle rotation. In the experiment on parameter change measurement before and after embedding, the algorithm is greatly reduced the volume of change in isotropic and fiber main direction of the myocardial fiber. In our proposed method, the main direction of the fiber is reduced by more than 100, and the average change of the mean diffusivity is reduced by more than 30 in the same database. In the visual quality of the algorithm, the peak signal-to-noise ratio is approximately 8 dB higher than that specified in the comparative literature. Conclusion An embedded selection feedback mechanism is proposed to carry out the selection of watermark embedding according to actual embedding demands. Then, the statistical histogram of the sub-band coefficient is constructed by specifying the fixed step length according to the characteristics of the wavelet transform coefficient of the DWI image. Finally, the reversible key algorithm based on the quantitative relationship between DWI and DTI is constructed. Experiments show that this algorithm can be applied to the watermark embedding of dispersion weighted imaging and can satisfy fiber direction as little as possible.
Key words
diffusion weight imaging (DWI); histogram; stable step; ration relationship; robust watermark
0 引言
弥散加权成像(DWI)[1-4]是通过施加多方向的弥散磁敏感梯度将每一弥散敏感梯度下组织中水分子的扩散运动转化为图像的灰度或其他参数信息的技术,该技术可以用于活体心脏心肌纤维建模[5]、大脑纤维、中枢神经系统病变、肝脏纤维化[6]等疾病的辅助诊断。随着远程医疗诊断技术的普及,使得越来越多的DWI数据应用于远程诊断和科学研究中,也让原本在医院单机上存储和使用的DWI图像需要通过网络进行传输和使用。为了防止未授权的DWI图像被使用或遭受篡改,提出对DWI数据进行保护具有重要的实用价值和研究意义。数字水印技术[7-8]作为保护数字媒体版权信息的一种重要技术,可以通过在DWI图像中加入特定版权标识信息从而增强DWI数据的信息安全性。虽然当前医学图像的版权保护通常采用可逆水印技术[9],但大多数可逆水印面对各种有意或无意的攻击时,几乎不具备鲁棒性。因此部分学者尝试研究具有鲁棒可逆性质的水印算法。
Zhang等人[10]提出将图像通过对冗余小波变换选取中频子带系数进行随机投影压缩,使用压缩矩阵相邻系数进行水印嵌入。该算法对部分几何攻击与信号攻击具有较好的鲁棒性,但需要携带观测矩阵使得需要传输的图像数据增大。Bekkouch等人[11]提出将图像的离散小波系数高频分量进行离散余弦变换,所获取的离散余弦系数采用奇异值分解,通过在奇异值矩阵中进行修改,从而完成水印的嵌入。该算法在旋转、裁剪、JPEG压缩、椒盐噪声均能较好地提取原始水印信息。An等人[12]提出使用属性启发式像素调整(PIPA)算法对子带系数进行调整,用于防止水印嵌入导致像素值溢出的情况。该算法在不同类型的图像中嵌入时,均具有较好的可视质量且提出使用聚类算法应用于水印算法的提取。
由于鲁棒可逆水印算法在提取水印信息时,通常需要携带观测矩阵,且图像在受到攻击时提取的图像无法保证无损可逆。因此使得一些医学图像信息保护领域的学者们开始探索和尝试将鲁棒零水印算法的思想应用于医学图像版权信息保护中。隋淼等人[13]提出一种基于离散余弦变换(DCT)的医学图像鲁棒零水印算法。该算法有效利用DCT低中频系数构建鲁棒零水印信息且具有较强的鲁棒性。但零水印[13-15]无法对图像的版权进行有效保护。
在与DWI成像领域相关学者讨论研究后得到如下的研究前提条件:在DWI计算转换得到的弥散张量成像(DTI)模型中,通过将计算参数改变范围有效地进行控制,从而保证计算得到的DTI模型对医生的诊断信息不会产生较为明显的差别。在此前提下本文提出将鲁棒水印算法应用于DWI医学图像版权信息保护中。本文提出一种基于DWI图像鲁棒水印算法。该算法首先使用最大类间方差分割算法和面积控制阈值将满足条件的切片作为待嵌入区域,并将该区域拟合为矩形,其次对待嵌入区域使用整数小波变换获取其低频子带系数,使用固定步长对低频子带系数进行直方图统计,并对系数直方图簇的比值关系进行调整从而嵌入鲁棒水印信息,最后对嵌入水印后图像的DWI表观系数与DTI弥散张量值比值关系作为可逆密钥。再一次对已嵌入图像进行加密,进一步增强本算法对DWI图像的版权信息保护。
1 DWI成像与DTI成像
1.1 DWI与DTI成像的发展和意义
1.2 常规MRI、DWI与DTI
常规MRI(magnetic resonance imaging)是一种以T1W1和T2W1为主的磁共振成像技术,主要显示人体器官或组织的形态结构及其信号强度变换。由于传统的MRI需要较长的成像时间,因此出现一些以快速成像序列为主的MRI成像技术,DWI成像技术就是其中一种。
DWI成像技术是对同一断层图像施加单个非零强度值的多方向弥散梯度,在不同方向弥散梯度下影响体内水分子流动并完成待检测部位影像采集。当强度值为零时,所获取的图像等同于传统MRI图像中的T2W1图像。
DTI图像通常采用多个梯度方向获取的DWI图像使用非线性最小二乘法建模计算得到图像中每个体素扩散张量,通过扩散张量进一步计算得到平均弥散程度
DWI和DTI的定义和计算过程在文献[15-18]中有详细介绍,本文不做过多赘述,现将本文要用到的关于DIW和DTI的各个参数和计算公式在表 1中进行说明。根据文献[15-18]相关计算,表 1列出DTI建模中将涉及的相关符号。
表 1
DTI建模过程中的相关变量、公式和定义
Table 1
Related variables, formulas, and definitions in the DTI modeling process
符号 | 含义 | 公式 |
张量矩阵在 |
||
张量矩阵中张量值 |
||
弥散梯度磁场强度 | ||
旋磁比 | ||
梯度时间间隔 | ||
3维空间中一点 |
||
弥散方向矢量矩阵 | ||
任意切片在不同扩散方向下灰度矩阵 | ||
张量矩阵 | ||
弥散方向下的平均信号量 | ||
|
||
张量矩阵特征值 | ||
张量矩阵特征向量 | ||
单个体素平均弥散程度 | ||
单个体素张量各向异性 | ||
主轴方向矢量 | ||
主轴方向与 |
||
主轴方向角改变量 |
本文所述算法通过将
2 DWI水印嵌入算法
2.1 最大可嵌入区域计算
最大类间方差分割法是一种基于全局求解最佳分割阈值的算法。假设图像为
$ V = {P_{\rm{f}}} \times {\left( {{I_{\rm{f}}} \leftarrow {I_{{\rm{arg}}}}} \right)^2} + {P_{\rm{b}}} \times {\left( {{I_{\rm{b}}} - {I_{{\rm{arg}}}}} \right)^2} $ | (1) |
$ {V_{\max }} = \arg \max (V) $ | (2) |
将每一切片在不同弥散方向下的图像通过最大类间方差分割法进行分割,从而提取出图像的前景区域,但提取的前景区域并不一定是规则形状,因此本文将前景区域内的所有像素点的
1) 本文读取一组患者DWI图像
2) 对步骤1)中所有的切片使用最大类间方差分割法提取扩散方向为
$ {P_{{\rm{ROI}}}} = \frac{{{S_{{\rm{ROI}}}}}}{{m \times n}} $ | (3) |
3) 获取所有切片多个扩散方向的前景区域像素点坐标位置集合
2.2 基于统计直方图修改的水印嵌入算法
灰度统计直方图是指通过统计一幅图像中所出现的灰度级个数,反映每个灰度级出现的频率。由于在DWI图像中
文献[20]中已证明,当图像遭受几何攻击时,虽然像素点的空间位置发生改变,但其统计直方图的主体形状不会发生较大改变。由离散小波变换基本原理可知,图像的低频子带系数是原图在最大尺度与最小分辨率下的最优逼近。因此低频子带系数的统计直方图对于几何攻击也应具有较好的几何不变性。
本文提出基于整数小波变换的嵌入算法,其具体步骤如下:
输入待嵌入DWI图像
输出嵌入后DWI图像
1) 使用随机函数生成一个二值混沌向量
2) 将DWI图像中的每一切片的不同扩散方向
$ \begin{array}{*{20}{c}} {{R_{\rm{c}}} = \left[ {(1 - \alpha )\arg avg \left( {\mathit{\boldsymbol{R}}_{LL}^{c,d}(u,v)} \right),} \right.}\\ {\left. {(1 + \alpha )\arg avg \left( {\mathit{\boldsymbol{R}}_{LL}^{c,d}(u,v)} \right)} \right]} \end{array} $ | (4) |
式中,
3) 将
当
当
4) 将修改后子带系数使用整数小波变换逆变换,生成已嵌入水印图
2.3 待调整系数选取规则
由表 1中公式可知,DWI图像在修改灰度值后对DTI求解到的结构参数有较大影响。为保证将嵌入后DWI图像所求得的DTI结构参数尽可能控制在一定范围内,本文提出一种基于标记的待调整系数选取规则。其具体步骤如下:
1) 将输入小波子带系数
2) 获取
3) 对修改后的子带系数进行重构,并计算是否存在有DTI角度参数
2.4 基于表观系数与弥散张量的可逆密钥设计
利用表 1中的公式,可对灰度值和张量矩阵之间的计算结果在特定情况下进行推导,因此本文算法对施加梯度矢量
$ S' = \left( {\begin{array}{*{20}{c}} {{S_1}}\\ \vdots \\ {2 \times {S_{12}}}\\ {2 \times {S_{13}}} \end{array}} \right) $ | (5) |
弥散梯度相关参数在采集时已经确定,即弥散梯度强度
$ \mathit{\boldsymbol{S'}} = \left( {\begin{array}{*{20}{c}} {{S_1}}\\ \vdots \\ {2 \times {S_{12}}\exp \left( {\sum\limits_{i = {x_{12}},{y_{12}},{z_{12}}} {\sum\limits_{j = {x_{12}},{y_{12}},{z_{12}}} {{b_{ij}}{{D'}_{ij}}} } } \right)}\\ {2 \times {S_{13}}\exp \left( {\sum\limits_{i = {x_{13}},{y_{13}},{z_{13}}} {\sum\limits_{j = {x_{13}},{y_{13}},{z_{13}}} {{b_{ij}}{{D'}_{ij}}} } } \right)} \end{array}} \right) $ | (6) |
由式(6)可知,不同弥散梯度方向下的灰度值发生改变,但
$ \ln \frac{{{S_1}}}{{{{S'}_b}}} = \sum\limits_{i \in {\mathit{\boldsymbol{D}}_L}} {\sum\limits_{j \in {\mathit{\boldsymbol{D}}_L}} {{b_{ij}}{{D'}_{ij}}} } ,{\mathit{\boldsymbol{D}}_L} = \left\{ {x,y,z} \right\} $ | (7) |
联立式(7)和表 1中公式,并代入式(5)中仅存在一个扩散方向时的值可得,当且仅当有一个扩散方向时其改变值与原值的关系如式(8)所示
$ \begin{array}{*{20}{c}} {\ln \left( {\frac{{{S_1}}}{{S_b^\prime }} - \ln \frac{{{S_1}}}{{{S_b}}}} \right) = \ln \left( {\frac{{{S_1}}}{{S_b^\prime }} \times \frac{{{S_b}}}{{{S_1}}}} \right) = \ln \left( {\frac{{{S_1}}}{{2 \times {S_b}}} \times \frac{{{S_b}}}{{{S_1}}}} \right) = }\\ { - \ln 2 + \ln 1 = \sum\limits_{i \in DL} {\sum\limits_{j \in DL} {\left( {{b_{ij}}D_{ij}^\prime - {b_{ij}}{D_{ij}}} \right)} } ,}\\ {{\mathit{\boldsymbol{D}}_L} = \{ x,y,z\} } \end{array} $ | (8) |
将其扩展至多方向时,由式(8)可推导出
$ \ln \frac{{{S_1}}}{{S_b^\prime }} = \left( {\begin{array}{*{20}{c}} {\ln \frac{{{S_1}}}{{{S_2}}} - \ln 2}\\ \vdots \\ {\ln \frac{{{S_1}}}{{{S_{12}}}} - \ln 2}\\ {\ln \frac{{{S_1}}}{{{S_{13}}}} - \ln 2} \end{array}} \right) = \left( {\begin{array}{*{20}{c}} {\ln \frac{{{S_1}}}{{{S_2}}}}\\ \vdots \\ {\ln \frac{{{S_1}}}{{{S_{12}}}}}\\ {\ln \frac{{{S_1}}}{{{S_{13}}}}} \end{array}} \right) - \left( {\begin{array}{*{20}{c}} {\ln 2}\\ \vdots \\ {\ln 2}\\ {\ln 2} \end{array}} \right) $ | (9) |
由于
2.5 DWI水印提取算法
2.5.1 可逆密钥解密
1) 利用面积控制阈值
2) 由步骤1获得嵌有可逆密钥与水印图像
$ \mathit{\boldsymbol{\widetilde I}}_f^{c,d} = \frac{{\widehat {\mathit{\boldsymbol{I}}_f^{c,d}}}}{{{k_n}}} $ | (10) |
2.5.2 鲁棒水印提取
1) 通过2.5.1节计算获得可逆密钥,使用该密钥可对加密后待提取水印信号的图像解密,获得图像
2)对步骤1)中获取到的
当相邻子带系数统计直方图之比
3) 由步骤2)获得
$ \mathit{\boldsymbol{w}}_{\left( k \right)}^{{\rm{end}}} = \left\{ \begin{array}{l} 1\;\;\;\;\;\sum\limits_{c = 1}^{nu{{m'}_c}} {{{w''}_{\left( {c,k} \right)}}} \ge \frac{{nu{{m'}_c}}}{2}\\ 0\;\;\;\;\;aaa \end{array} \right. $ | (11) |
在本文中使用在不同切片中反复重嵌入相同水印值,从而有效降低提取水印时误判率,提高提取后水印的鲁棒性。
3 仿真实验及分析
本文使用MATLAB 2017b作为实验平台进行实验,本实验室于不同设备中采集到约42位病人,每位病人采集25~54个切层,采集方向数量为13~65个方向,共计约20 000余幅图像。在DWI图像数据库中任选一组病人的DWI图像用于实验结果的说明。通过多次实验后, 本文设置面积阈值
3.1 主方向变化夹角
在DTI成像中,主轴方向反映体素的纤维走向是医生用于诊断纤维病变的关键依据,因此保证纤维主方向变化夹角在一定范围内是本文算法中对于医生诊断不影响的最大保证。使用表 1中公式计算原始主轴方向矢量
表 2
不同切片嵌入后参数量改变统计
Table 2
Statistics on the number of parameters changed after embedding in different slices
参数名称 | 改变量/总量 |
1/4 899 | |
4/4 899 | |
0/4 899 |
由表 2可知,利用本文所述算法,对33个切层进行水印嵌入后,较少出现像素点嵌入前后夹角
3.2 实验前后各参数统计情况
任意选取一组图像中某一切层弥散加权图像对各参数进行计算, 如图 4所示。图 4(a)(d)(g)为未嵌入图像前的各参数热图,图 4(b)(e)(h)为嵌入水印后各参数热图,图 4(c)(f)(i)为嵌入前后各参数变化量绘制的对应热图。由图 4可知,各参数虽存在少量变化,但没有超出临床医生所要求的标准,且出现位置皆在非心肌纤维处,因此本算法可以适用于临床。
3.3 评价指标
为评价本文所提算法的性能,本文将从视觉质量和水印信息鲁棒性两个指标对算法性能进行客观分析和测量。
3.3.1 视觉质量评价
峰值信噪比(PSNR)是一种通过客观统计进行水印嵌入前后图像质量标准的方法。当嵌入前后的图像的差异越小时,PSNR的值越大。通常认为当PSNR大于35 dB时图像的差异通过肉眼已经无法进行判别。
$ PSNR = 10{\rm{lg}}\left( {\frac{{m \times n \times {2^{12}} \times {2^{12}}}}{{\sum\limits_{i = 0}^{m - 1} {\sum\limits_{j = 0}^{n - 1} {{{\left( {\mathit{\boldsymbol{\tilde I}}_f^{c,d}(i,j) - \mathit{\boldsymbol{I}}_f^{c,d}(i,j)} \right)}^2}} } }}} \right) $ | (12) |
3.3.2 归一化相关系数
归一化相关系数(NC)用于判断在嵌入图像遭受攻击前后水印信号的差别,以反映水印嵌入算法的鲁棒性,具体为
$ NC = \frac{{\sum\limits_{i = 1}^{nw} {{{\mathit{\boldsymbol{W'}}}_{\left( k \right)}}\left( i \right)} \times {\mathit{\boldsymbol{w}}^{{\rm{end}}}}\left( i \right)}}{{\sum\limits_{i - 1}^{nw} {{{\left( {{{\mathit{\boldsymbol{W'}}}_{\left( k \right)}}\left( i \right)} \right)}^2}} }} $ | (13) |
3.4 水印鲁棒性分析
3.4.1 旋转攻击
旋转攻击是在图像攻击中一种常见的几何攻击方式,它利用旋转变换,改变原有灰度所处的空间位置,其中旋转变换分为旋转变换和带裁剪的旋转变换。其中带裁剪的变换是指由于对原始图像进行旋转攻击后图像尺寸发生变化,因此为保证旋转后图像尺寸与原始图像尺寸保持一致,对旋转后图像进行边缘裁剪。在本文所提算法的实验中,当旋转角度在1°~ 5°时,
表 3
DWI图像在带裁剪的旋转攻击下的NC值
Table 3
NC value of DWI with cropped rotation attack
带裁剪的旋转攻击方式 | NC |
带裁剪的旋转1° | 1 |
旋转1° | 1 |
带裁剪的旋转2° | 1 |
旋转2° | 1 |
带裁剪的旋转5° | 1 |
旋转5° | 1 |
带裁剪的旋转10° | 1 |
旋转10° | 1 |
带裁剪的旋转15° | 0.98 |
旋转15° | 1 |
3.4.2 缩放攻击
3.5 噪声攻击
噪声攻击是指在原始图像信号中加入一定的噪声信号,使得原始图像信号遭到篡改。本文使用椒盐噪声和高斯噪声作为主要的噪声攻击方式,本文选择对所有切层下所有方向图片分别加入一定量的高斯噪声和椒盐噪声。实验结果如表 5所示。通过实验可知,本文所提算法在不同噪声攻击下的
表 5
DWI图像在噪声攻击下的NC值
Table 5
NC value of DWI under noise attack
噪声方差 | NC | |
椒盐噪声 | 0.01 | 1 |
0.02 | 1 | |
0.05 | 0.91 | |
0.1 | 0.90 | |
高斯噪声 | 0.02 | 0.82 |
0.05 | 0.79 | |
0.1 | 0.79 |
3.6 低通滤波
通常在远程医疗传输接收端为减少信道中存在高斯噪声对医生诊断的影响会使用低通滤波器对图像进行预处理,因此在DWI的水印中能抵御低通滤波有助于在图像传输后在接收端依旧可以提取版权信息。本算法在进行高斯低通滤波攻击时,NC值为1,因此本文所提算法针对高斯低通滤波具有很强的鲁棒性。
3.7 信号增强攻击
3.8 对比实验
本文选取切片数量为54层,施加弥散梯度方向数为12,大小为128×128像素的人体弥散加权心脏图像作为测试图像,但当前尚无针对弥散加权成像的不可见鲁棒水印算法。本文选取文献[20]的鲁棒水印算法作为对比实验组,将该算法应用于本文所使用的弥散加权图像中,对比两种算法在弥散加权成像中的结果。
3.8.1 角度变化对比实验
本文首次将鲁棒水印应用于DWI图像的版权保护中。(为此本文提出相关算法可适用于弥散加权成像的前程):需要使得嵌入水印后的图像不仅满足在视觉上使得人眼无法感知,还需要使得嵌入后DWI建模形成的DTI参数中相应纤维主方向变化度数
将本文算法与文献[20]中提出的低频水印算法进行对比。图 5(a)(c)(e)为原始切层图像在
3.8.2 视觉质量对比实验
将本算法与文献[20]算法对心脏非心尖、心底区域所在的33个切片,先计算每一个切片的PSNR值,然后再求平均得到的整个嵌入切片的平均PSNR值进行比较,其结果如表 7所示。
表 7
视觉评价指标比较
Table 7
Comparison of visual evaluation indexes
/dB | ||
文献[20] | 本文算法 | |
PSNR | 76.658 | 84.181 6 |
根据反复实验,在该情况下,水印信号无法被人眼直接察觉,但从表 8可知,在同样的切片数和方向数下,本文所提算法在视觉评价指标中明显好于文献[20]中的算法。
表 8
鲁棒性对比情况
Table 8
Robust contrast situation
攻击 | NC值 | |
文献[20] | 本文算法 | |
无 | 1 | 1 |
带旋转裁剪1° | 1 | 1 |
带旋转裁剪2° | 1 | 0.98 |
带旋转裁剪5° | 0.98 | 1 |
带旋转裁剪15° | 0.98 | 0.98 |
缩放0.7 | 0.62 | 0.78 |
缩放0.9 | 0.71 | 0.69 |
缩放1.2 | 0.81 | 0.81 |
直方图均衡化 | 0.70 | 0.83 |
高斯滤波 | 1 | 1 |
椒盐噪声0.01 | 1 | 1 |
椒盐噪声0.02 | 1 | 1 |
高斯噪声0.01 | 0.71 | 0.82 |
高斯噪声0.02 | 0.68 | 0.82 |
旋转1° | 0.84 | 1 |
旋转5° | 1 | 1 |
旋转10° | 1 | 1 |
对比度扩展10% | 0.72 | 0.83 |
3.8.3 鲁棒性对比实验
3.8.4 嵌入鲁棒水印参数改变量对比
4 结论
本文提出一种基于直方图统计修改的弥散加权图像鲁棒水印。该算法首先利用最大类间方差法提取患者所有切片中的前景区域,并使用面积阈值选取待嵌入切片。其次将待嵌入切片中所有弥散方向前景区域图像进行整数小波变换,对指定范围内的低频小波区间使用固定步长统计子带系数直方图。然后调整相邻系数统计直方图簇的比值关系完成水印信息的嵌入。最后使用表观系数和弥散张量的关联,设计可逆密钥。该算法在图像遭受几何攻击和噪声攻击时,具有一定鲁棒性且该算法在生成的DTI角度参数中改变较小,因此本算法适用于DWI图像水印嵌入。且本算法在与参考文献[20]所提出的算法进行对比后,在鲁棒性、角度参数和各项同性参数上具有较明显的优势,能够有效控制DTI参数,且本文通过二次嵌入使得未获得密钥时也无法正确提取相应信息,从而增强图像安全性。在本文中对于参数的控制只能通过反馈网络进行,为自适应地进行修改,将提出使用神经网络等算法进行修改。
附录 DWI表观系数与DTI张量推导关系
已知原始DWI表观系数为
$ \mathit{\boldsymbol{S}} = \left( {\begin{array}{*{20}{c}} {{S_1}}\\ \vdots \\ {{S_{N - 1}}}\\ {{S_N}} \end{array}} \right) $ | (14) |
修改系数变量
$ \mathit{\boldsymbol{S'}} = \left( {\begin{array}{*{20}{c}} {{S_1}}\\ \vdots \\ {v \times {S_{N - 1}}}\\ {v \times {S_N}} \end{array}} \right) $ | (15) |
根据文献[10]中所述,使用表 1中公式运用加权线性最小二乘法对矩阵
其中,修改表观系数后为
$ {\mathit{\boldsymbol{ \boldsymbol{\varSigma} }}^{ - 1}} = diag \left( {\frac{{{{\left( {v \times {S_i}} \right)}^2}}}{{\sigma _i^2}}} \right) $ | (16) |
因此修改后矩阵
$ {\mathit{\boldsymbol{B}}^{\rm{T}}}{\mathit{\boldsymbol{ \boldsymbol{\varSigma} }}^{ - 1}}\mathit{\boldsymbol{B}} =\\ \left( {\begin{array}{*{20}{c}} {\sum\limits_{i = 1}^N {b_{xxi}^2} \frac{{{{\left( {S_i^\prime } \right)}^2}}}{{\sigma _i^2}}}&{\sum\limits_{i = 1}^N {{b_{xxi}}} {b_{yyi}}\frac{{{{\left( {S_i^\prime } \right)}^2}}}{{\sigma _i^2}}}& \cdots &{2\sum\limits_{i = 1}^N {{b_{xxi}}} {b_{yzi}}\frac{{{{\left( {S_i^\prime } \right)}^2}}}{{\sigma _i^2}}}&{ - \sum\limits_{i = 1}^N {{b_{xxi}}} \frac{{{{\left( {S_i^\prime } \right)}^2}}}{{\sigma _i^2}}}\\ {\sum\limits_{i = 1}^N {{b_{yyi}}} {b_{xxi}}\frac{{{{\left( {S_i^\prime } \right)}^2}}}{{\sigma _i^2}}}&{\sum\limits_{i = 1}^N {b_{yyi}^2} \frac{{{{\left( {S_i^\prime } \right)}^2}}}{{\sigma _i^2}}}& \cdots &{2\sum\limits_{i = 1}^N {{b_{yyi}}} {b_{yzi}}\frac{{{{\left( {S_i^\prime } \right)}^2}}}{{\sigma _i^2}}}&{ - \sum\limits_{i = 1}^N {{b_{yyi}}} \frac{{{{\left( {S_i^\prime } \right)}^2}}}{{\sigma _i^2}}}\\ {\sum\limits_{i = 1}^N {{b_{zzi}}} {b_{xxi}}\frac{{{{\left( {S_i^\prime } \right)}^2}}}{{\sigma _i^2}}}&{\sum\limits_{i = 1}^N {{b_{zzi}}} {b_{yyi}}\frac{{{{\left( {S_i^\prime } \right)}^2}}}{{\sigma _i^2}}}& \cdots &{2\sum\limits_{i = 1}^N {{b_{zzi}}} {b_{yzi}}\frac{{{{\left( {S_i^\prime } \right)}^2}}}{{\sigma _i^2}}}&{ - \sum\limits_{i = 1}^N {{b_{zzi}}} \frac{{{{\left( {S_i^\prime } \right)}^2}}}{{\sigma _i^2}}}\\ {2\sum\limits_{i = 1}^N {{b_{xyi}}} {b_{yzi}}\frac{{{{\left( {S_i^\prime } \right)}^2}}}{{\sigma _i^2}}}&{2\sum\limits_{i = 1}^N {{b_{xyi}}} {b_{yyi}}\frac{{{{\left( {S_i^\prime } \right)}^2}}}{{\sigma _i^2}}}& \cdots &{4\sum\limits_{i = 1}^N {{b_{xyi}}} {b_{yzi}}\frac{{{{\left( {S_i^\prime } \right)}^2}}}{{\sigma _i^2}}}&{ - 2\sum\limits_{i = 1}^N {{b_{xyi}}} \frac{{{{\left( {S_i^\prime } \right)}^2}}}{{\sigma _i^2}}}\\ {2\sum\limits_{i = 1}^N {{b_{xzi}}} {b_{xxi}}\frac{{{{\left( {S_i^\prime } \right)}^2}}}{{\sigma _i^2}}}&{2\sum\limits_{i = 1}^N {{b_{xzi}}} {b_{yyi}}\frac{{{{\left( {S_i^\prime } \right)}^2}}}{{\sigma _i^2}}}& \cdots &{4\sum\limits_{i = 1}^N {{b_{xzi}}} {b_{yzi}}\frac{{{{\left( {S_i^\prime } \right)}^2}}}{{\sigma _i^2}}}&{ - 2\sum\limits_{i = 1}^N {{b_{xzi}}} \frac{{{{\left( {S_i^\prime } \right)}^2}}}{{\sigma _i^2}}}\\ {2\sum\limits_{i = 1}^N {{b_{yzi}}} {b_{xxi}}\frac{{{{\left( {S_i^\prime } \right)}^2}}}{{\sigma _i^2}}}&{2\sum\limits_{i = 1}^N {{b_{yzi}}} {b_{yyi}}\frac{{{{\left( {S_i^\prime } \right)}^2}}}{{\sigma _i^2}}}& \cdots &{4\sum\limits_{i = 1}^N {b_{yzi}^2} \frac{{{{\left( {S_i^\prime } \right)}^2}}}{{\sigma _i^2}}}&{ - 2\sum\limits_{i = 1}^N {{b_{yzi}}} \frac{{{{\left( {S_i^\prime } \right)}^2}}}{{\sigma _i^2}}}\\ { - \sum\limits_{i = 1}^N {{b_{xxi}}} \frac{{{{\left( {S_i^\prime } \right)}^2}}}{{\sigma _i^2}}}&{ - \sum\limits_{i = 1}^N {{b_{yyi}}} \frac{{{{\left( {S_i^\prime } \right)}^2}}}{{\sigma _i^2}}}& \cdots &{ - 2\sum\limits_{i = 1}^N {{b_{yzi}}} \frac{{{{\left( {S_i^\prime } \right)}^2}}}{{\sigma _i^2}}}&{ - \sum\limits_{i = 1}^N {\frac{{{{\left( {S_i^\prime } \right)}^2}}}{{\sigma _i^2}}} } \end{array}} \right) $ | (17) |
$ \mathit{\boldsymbol{x'}} = \left( {\begin{array}{*{20}{c}} {\ln \left( {{S_1}} \right)}\\ \vdots \\ {\ln \left( {v \times {S_{N - 1}}} \right)}\\ {\ln \left( {v \times {S_N}} \right)} \end{array}} \right) $ | (18) |
因此联立式(17)和式(18)可知
$ \begin{array}{*{20}{c}} {{{\left( {{\mathit{\boldsymbol{B}}^{\rm{T}}}{\mathit{\boldsymbol{ \boldsymbol{\varSigma} }}^{ - 1}}\mathit{\boldsymbol{B}}} \right)}^{ - 1}}\left( {{\mathit{\boldsymbol{B}}^{\rm{T}}}{\mathit{\boldsymbol{ \boldsymbol{\varSigma} }}^{ - 1}}} \right)\mathit{\boldsymbol{x'}} = }\\ {\left( {\begin{array}{*{20}{c}} {\sum\limits_{i = 1}^N {b_{xxi}^2} \frac{{{{\left( {S_i^\prime } \right)}^2}}}{{\sigma _i^2}}}&{\sum\limits_{i = 1}^N {{b_{xxi}}} {b_{yyi}}\frac{{{{\left( {S_i^\prime } \right)}^2}}}{{\sigma _i^2}}}& \cdots &{2\sum\limits_{i = 1}^N {{b_{xxi}}} {b_{yzi}}\frac{{{{\left( {S_i^\prime } \right)}^2}}}{{\sigma _i^2}}}&{ - \sum\limits_{i = 1}^N {{b_{xxi}}} \frac{{{{\left( {S_i^\prime } \right)}^2}}}{{\sigma _i^2}}}\\ {\sum\limits_{i = 1}^N {{b_{yyi}}} {b_{xxi}}\frac{{{{\left( {S_i^\prime } \right)}^2}}}{{\sigma _i^2}}}&{\sum\limits_{i = 1}^N {b_{yyi}^2} \frac{{{{\left( {S_i^\prime } \right)}^2}}}{{\sigma _i^2}}}& \cdots &{2\sum\limits_{i = 1}^N {{b_{yyi}}} {b_{yzi}}\frac{{{{\left( {S_i^\prime } \right)}^2}}}{{\sigma _i^2}}}&{ - \sum\limits_{i = 1}^N {{b_{yyi}}} \frac{{{{\left( {S_i^\prime } \right)}^2}}}{{\sigma _i^2}}}\\ {\sum\limits_{i = 1}^N {{b_{zzi}}} {b_{xxi}}\frac{{{{\left( {S_i^\prime } \right)}^2}}}{{\sigma _i^2}}}&{\sum\limits_{i = 1}^N {{b_{zzi}}} {b_{yyi}}\frac{{{{\left( {S_i^\prime } \right)}^2}}}{{\sigma _i^2}}}& \cdots &{2\sum\limits_{i = 1}^N {{b_{zzi}}} {b_{yzi}}\frac{{{{\left( {S_i^\prime } \right)}^2}}}{{\sigma _i^2}}}&{ - \sum\limits_{i = 1}^N {{b_{zzi}}} \frac{{{{\left( {S_i^\prime } \right)}^2}}}{{\sigma _i^2}}}\\ {2\sum\limits_{i = 1}^N {{b_{xyi}}} {b_{yzi}}\frac{{{{\left( {S_i^\prime } \right)}^2}}}{{\sigma _i^2}}}&{2\sum\limits_{i = 1}^N {{b_{xyi}}} {b_{yyi}}\frac{{{{\left( {S_i^\prime } \right)}^2}}}{{\sigma _i^2}}}& \cdots &{4\sum\limits_{i = 1}^N {{b_{xyi}}} {b_{yzi}}\frac{{{{\left( {S_i^\prime } \right)}^2}}}{{\sigma _i^2}}}&{ - 2\sum\limits_{i = 1}^N {{b_{xyi}}} \frac{{{{\left( {S_i^\prime } \right)}^2}}}{{\sigma _i^2}}}\\ {2\sum\limits_{i = 1}^N {{b_{xzi}}} {b_{xxi}}\frac{{{{\left( {S_i^\prime } \right)}^2}}}{{\sigma _i^2}}}&{2\sum\limits_{i = 1}^N {{b_{xzi}}} {b_{yyi}}\frac{{{{\left( {S_i^\prime } \right)}^2}}}{{\sigma _i^2}}}& \cdots &{4\sum\limits_{i = 1}^N {{b_{xzi}}} {b_{yzi}}\frac{{{{\left( {S_i^\prime } \right)}^2}}}{{\sigma _i^2}}}&{ - 2\sum\limits_{i = 1}^N {{b_{xzi}}} \frac{{{{\left( {S_i^\prime } \right)}^2}}}{{\sigma _i^2}}}\\ {2\sum\limits_{i = 1}^N {{b_{yzi}}} {b_{xxi}}\frac{{{{\left( {S_i^\prime } \right)}^2}}}{{\sigma _i^2}}}&{2\sum\limits_{i = 1}^N {{b_{yzi}}} {b_{yyi}}\frac{{{{\left( {S_i^\prime } \right)}^2}}}{{\sigma _i^2}}}& \cdots &{4\sum\limits_{i = 1}^N {b_{yzi}^2} \frac{{{{\left( {S_i^\prime } \right)}^2}}}{{\sigma _i^2}}}&{ - 2\sum\limits_{i = 1}^N {{b_{yzi}}} \frac{{{{\left( {S_i^\prime } \right)}^2}}}{{\sigma _i^2}}}\\ { - \sum\limits_{i = 1}^N {{b_{xxi}}} \frac{{{{\left( {S_i^\prime } \right)}^2}}}{{\sigma _i^2}}}&{ - \sum\limits_{i = 1}^N {{b_{yyi}}} \frac{{{{\left( {S_i^\prime } \right)}^2}}}{{\sigma _i^2}}}& \cdots &{ - 2\sum\limits_{i = 1}^N {{b_{yzi}}} \frac{{{{\left( {S_i^\prime } \right)}^2}}}{{\sigma _i^2}}}&{ - \sum\limits_{i = 1}^N {\frac{{{{\left( {S_i^\prime } \right)}^2}}}{{\sigma _i^2}}} } \end{array}} \right)}\\ {\left( {\begin{array}{*{20}{c}} { - {b_{xx1}}\frac{{S_1^2}}{{\sigma _1^2}}}&{ - {b_{xx2}}\frac{{{{\left( {v \times {S_2}} \right)}^2}}}{{\sigma _2^2}}}& \cdots &{ - {b_{xxN}}\frac{{{{\left( {v \times {S_N}} \right)}^2}}}{{\sigma _N^2}}}\\ { - {b_{yy1}}\frac{{S_1^2}}{{\sigma _1^2}}}&{ - {b_{yy2}}\frac{{{{\left( {v \times {S_2}} \right)}^2}}}{{\sigma _2^2}}}& \cdots &{ - {b_{{\rm{yyN}}}}\frac{{{{\left( {v \times {S_N}} \right)}^2}}}{{\sigma _N^2}}}\\ \vdots & \vdots & \cdots &{}\\ { - 2{b_{{\rm{xy1}}}}\frac{{S_1^2}}{{\sigma _1^2}}}&{ - 2{b_{{\rm{xy2}}}}\frac{{{{\left( {v \times {S_2}} \right)}^2}}}{{\sigma _2^2}}}& \cdots &{ - 2{b_{{\rm{xyN}}}}\frac{{{{\left( {v \times {S_N}} \right)}^2}}}{{\sigma _N^2}}}\\ { - 2{b_{{\rm{xz1}}}}\frac{{S_1^2}}{{\sigma _1^2}}}&{ - 2{b_{{\rm{xz2}}}}\frac{{{{\left( {v \times {S_2}} \right)}^2}}}{{\sigma _2^2}}}& \cdots &{ - 2{b_{{\rm{xzN}}}}\frac{{{{\left( {v \times {S_N}} \right)}^2}}}{{\sigma _N^2}}}\\ {\frac{{S_1^2}}{{\sigma _1^2}}}&{\frac{{{{\left( {v \times {S_2}} \right)}^2}}}{{\sigma _2^2}}}& \cdots &{\frac{{{{\left( {v \times {S_N}} \right)}^2}}}{{\sigma _N^2}}} \end{array}} \right)} \end{array} $ | (19) |
$ {\mathit{\boldsymbol{B}}^{\rm{T}}}\mathit{\boldsymbol{B}} = \left( {\begin{array}{*{20}{c}} { - {b_{xxi}}}& \cdots &{ - {b_{xxN}}}\\ { - {b_{yyi}}}& \cdots &{ - {b_{yyN}}}\\ { - {b_{zzi}}}& \cdots &{ - {b_{zzN}}}\\ { - {b_{xyi}}}& \cdots &{ - {b_{xyN}}}\\ { - {b_{xzi}}}& \cdots &{ - {b_{xzN}}}\\ { - {b_{yzi}}}& \cdots &{ - {b_{yzN}}}\\ 1& \cdots &1 \end{array}} \right)\left( {\begin{array}{*{20}{c}} { - {b_{xxi}}}&{ - {b_{yyi}}}&{ - {b_{zzi}}}&{ - 2{b_{xyi}}}&{ - 2{b_{xzi}}}&{ - 2{b_{yzi}}}&1\\ \vdots & \vdots & \vdots & \vdots & \vdots & \vdots & \vdots \\ { - {b_{xxN}}}&{ - {b_{yyN}}}&{ - {b_{zzN}}}&{ - 2{b_{xyN}}}&{ - 2{b_{xzN}}}&{ - 2{b_{yzN}}}&1 \end{array}} \right) $ | (20) |
$ \mathit{\boldsymbol{M}} = \left( {\begin{array}{*{20}{c}} {\begin{array}{*{20}{c}} {\frac{{S_1^2}}{{\sigma _1^2}}}&0& \cdots &0 \end{array}}\\ {\begin{array}{*{20}{c}} 0&{\frac{{S_2^2}}{{\sigma _2^2}}}& \cdots &0 \end{array}}\\ \vdots \\ {\begin{array}{*{20}{c}} 0&0&0&{\frac{{S_N^2}}{{\sigma _N^2}}} \end{array}} \end{array}} \right) $ | (21) |
$ {\mathit{\boldsymbol{M}}_1} = \mathit{\boldsymbol{M}} * \left[ {\begin{array}{*{20}{c}} {\begin{array}{*{20}{c}} 1& \cdots &0&0 \end{array}}\\ {\begin{array}{*{20}{c}} 0&{{v^2}}& \cdots &0 \end{array}}\\ \vdots \\ {\begin{array}{*{20}{c}} 0&0&0&{{v^2}} \end{array}} \end{array}} \right] $ | (22) |
$ \left( {{\mathit{\boldsymbol{B}}^{\rm{T}}}{\mathit{\Sigma }^{ - 1}}\mathit{\boldsymbol{B}}} \right) = {\mathit{\boldsymbol{B}}^{\rm{T}}} \times \left( {\mathit{\boldsymbol{M}} * \left[ {\begin{array}{*{20}{c}} {\begin{array}{*{20}{c}} 1& \cdots &0&0 \end{array}}\\ {\begin{array}{*{20}{c}} 0&{{v^2}}& \cdots &0 \end{array}}\\ \vdots \\ {\begin{array}{*{20}{c}} 0&0&0&{{v^2}} \end{array}} \end{array}} \right]} \right) \times \mathit{\boldsymbol{B}} $ | (23) |
根据三重积公式可得
$ \left( {{\mathit{\boldsymbol{B}}^{\rm{T}}}{\mathit{\Sigma }^{ - 1}}\mathit{\boldsymbol{B}}} \right) = \left[ {\begin{array}{*{20}{c}} {\begin{array}{*{20}{c}} 1& \cdots &0&0 \end{array}}\\ {\begin{array}{*{20}{c}} 0&{{v^2}}& \cdots &0 \end{array}}\\ \vdots \\ {\begin{array}{*{20}{c}} 0&0&0&{{v^2}} \end{array}} \end{array}} \right]*\left( {{\mathit{\boldsymbol{B}}^{\rm{T}}}\mathit{\boldsymbol{MB}}} \right) $ | (24) |
对该矩阵求逆可得
$ {{{\left( {{\mathit{\boldsymbol{B}}^{\rm{T}}}{\mathit{\Sigma }^{ - 1}}\mathit{\boldsymbol{B}}} \right)}^{ - 1}} = {{\left[ {\begin{array}{*{20}{c}} {\begin{array}{*{20}{c}} 1& \cdots &0&0 \end{array}}\\ {\begin{array}{*{20}{c}} 0&{{v^2}}& \cdots &0 \end{array}}\\ \vdots \\ {\begin{array}{*{20}{c}} 0&0&0&{{v^2}} \end{array}} \end{array}} \right]}^{ - 1}} * {{\left( {{\mathit{\boldsymbol{B}}^{\rm{T}}}\mathit{\boldsymbol{MB}}} \right)}^{ - 1}}} $ | (25) |
$ \begin{array}{*{20}{c}} {{\mathit{\boldsymbol{B}}^{\rm{T}}}{\mathit{\Sigma }^{ - 1}} = \left( {\begin{array}{*{20}{c}} 1&{{v^2}}&{{v^2}}&{{v^2}}\\ 1&{{v^2}}&{{v^2}}&{{v^2}}\\ 1&{{v^2}}&{{v^2}}&{{v^2}}\\ 1&{{v^2}}&{{v^2}}&{{v^2}} \end{array}} \right) * }\\ {\left( {\begin{array}{*{20}{c}} { - {b_{xx1}}\frac{{S_1^2}}{{\sigma _1^2}}}&{ - {b_{xx2}}\frac{{{{\left( {{S_2}} \right)}^2}}}{{\sigma _2^2}}}& \cdots &{ - {b_{xxN}}\frac{{{{\left( {{S_N}} \right)}^2}}}{{\sigma _N^2}}}\\ { - {b_{yy1}}\frac{{S_1^2}}{{\sigma _1^2}}}&{ - {b_{yy2}}\frac{{{{\left( {{S_2}} \right)}^2}}}{{\sigma _2^2}}}& \cdots &{ - {b_{yyN}}\frac{{{{\left( {{S_N}} \right)}^2}}}{{\sigma _N^2}}}\\ \vdots & \vdots & \vdots & \vdots \\ { - 2{b_{xy1}}\frac{{S_1^2}}{{\sigma _1^2}}}&{ - 2{b_{xy2}}\frac{{{{\left( {{S_2}} \right)}^2}}}{{\sigma _2^2}}}& \cdots &{ - 2{b_{xyN}}\frac{{{{\left( {{S_N}} \right)}^2}}}{{\sigma _N^2}}}\\ { - 2{b_{xz1}}\frac{{S_1^2}}{{\sigma _1^2}}}&{ - 2{b_{xz2}}\frac{{{{\left( {{S_2}} \right)}^2}}}{{\sigma _2^2}}}& \cdots &{ - 2{b_{xzN}}\frac{{{{\left( {{S_N}} \right)}^2}}}{{\sigma _N^2}}}\\ {\frac{{S_1^2}}{{\sigma _1^2}}}&{\frac{{{{\left( {{S_2}} \right)}^2}}}{{\sigma _2^2}}}& \cdots &{\frac{{{{\left( {{S_N}} \right)}^2}}}{{\sigma _N^2}}} \end{array}} \right)} \end{array} $ | (26) |
首先求解实对称矩阵的逆矩阵即求
再次使用三重积性质可得
$ {\left( {{\mathit{\boldsymbol{B}}^{\rm{T}}}{{\mathit{\Sigma '}}^{ - 1}}\mathit{\boldsymbol{B}}} \right)^{ - 1}}\left( {{\mathit{\boldsymbol{B}}^{\rm{T}}}{{\mathit{\Sigma '}}^{ - 1}}} \right) = {\left( {{\mathit{\boldsymbol{B}}^{\rm{T}}}{\mathit{\Sigma }^{ - 1}}\mathit{\boldsymbol{B}}} \right)^{ - 1}}\left( {{\mathit{\boldsymbol{B}}^{\rm{T}}}{\mathit{\Sigma }^{ - 1}}} \right) $ | (27) |
因此
$ \begin{array}{*{20}{c}} {\alpha ' - \alpha = {{\left( {{\mathit{\boldsymbol{B}}^{\rm{T}}}{\mathit{\Sigma }^{ - 1}}\mathit{\boldsymbol{B}}} \right)}^{ - 1}}\left( {{\mathit{\boldsymbol{B}}^{\rm{T}}}{\mathit{\Sigma }^{ - 1}}} \right)x' - }\\ {{{\left( {{\mathit{\boldsymbol{B}}^{\rm{T}}}{\mathit{\Sigma }^{ - 1}}\mathit{\boldsymbol{B}}} \right)}^{ - 1}}\left( {{\mathit{\boldsymbol{B}}^{\rm{T}}}{\mathit{\Sigma }^{ - 1}}} \right)\mathit{\boldsymbol{x}}} \end{array} $ | (28) |
又因为
$ \mathit{\boldsymbol{x'}} = \left( {\begin{array}{*{20}{c}} {\ln \left( {{S_1}} \right)}\\ {\ln \left( {v \times {S_2}} \right)}\\ \vdots \\ {\ln \left( {v \times {S_N}} \right)} \end{array}} \right) = \left( {\begin{array}{*{20}{c}} {\ln \left( {{S_1}} \right)}\\ {\ln \left( {{S_2}} \right)}\\ \vdots \\ {\ln \left( {{S_N}} \right)} \end{array}} \right) + \left( {\begin{array}{*{20}{c}} {\ln 1}\\ {\ln v}\\ \vdots \\ {\ln v} \end{array}} \right) $ | (29) |
综上所述,当已知变换倍数时,可以求得原状态下的张量矩阵。
参考文献
-
[1] Hao D P, Xu W J, Xu A D. Basic principles of MR dispersion weighted imaging and progress in clinical application[J]. Chinese Imaging Journal of Integrated Traditional and Western Medicine, 2003, 1(4): 244–247. [郝大鹏, 徐文坚, 徐爱德. MR弥散加权成像基本原理及临床应用进展[J]. 中国中西医结合影像学杂志, 2003, 1(4): 244–247. ] [DOI:10.3969/j.issn.1672-0512.2003.04.021]
-
[2] Bammer R. Basic principles of diffusion-weighted imaging[J]. European Journal of Radiology, 2003, 45(3): 169–184. [DOI:10.1016/S0720-048X(02)00303-0]
-
[3] Li Z W, Xia L M. Preliminary study on the dispersion weighted imaging of myocardial magnetic resonance multi-B value[C]//Proceedings of the 15th National Cardiovascular Disease Conference of the Middle Hua Medical Society. Tianjin, China: Chinese Medical Association, 2013. [ [李志伟, 夏黎明.心肌磁共振多b值弥散加权成像的初步探讨[C]//中华医学会第十五次全国心血管病学大会论文汇编.天津: 中华医学会, 2013.]
-
[4] Liu H. The basic principle and clinical application value of MR whole body diffusion-weighted imaging[J]. Journal of Medical Imaging, 2013, 23(7): 1126–1129. [刘辉. 磁共振全身弥散加权成像的基本原理及临床应用价值[J]. 医学影像学杂志, 2013, 23(7): 1126–1129. ] [DOI:10.3969/j.issn.1006-9011.2013.07.046]
-
[5] Wang Q S, Guo Q Y, Liang C H, et al. Diffusion weighted MR imaging in experimental liver fibrosis model in rabbits: preliminary study[J]. Chinese Journal of Medical Imaging Technology, 2007, 23(7): 952–955. [王秋实, 郭启勇, 梁长虹, 等. MR弥散加权成像在兔肝纤维化模型中的初步实验研究[J]. 中国医学影像技术, 2007, 23(7): 952–955. ] [DOI:10.3321/j.issn:1003-3289.2007.07.003]
-
[6] Mildenberger P, Eichelberg M, Martin E. Introduction to the DICOM standard[J]. European Radiology, 2002, 12(4): 920–927. [DOI:10.1007/s003300101100]
-
[7] Sun S H, Lu Z M. Digital watermarking techniques[J]. Acta Electronica Sinica, 2000, 28(8): 85–90. [孙圣和, 陆哲明. 数字水印处理技术[J]. 电子学报, 2000, 28(8): 85–90. ] [DOI:10.3321/j.issn:0372-2112.2000.08.025]
-
[8] Yi K X, Shi J Y, Sun X. Digital watermarking techniques: an introductory review[J]. Journal of Image and Graphics, 2001, 6(2): 111–117. [易开祥, 石教英, 孙鑫. 数字水印技术研究进展[J]. 中国图象图形学报, 2001, 6(2): 111–117. ] [DOI:10.11834/jig.20010229]
-
[9] Deng X H, Chen Z G, Mao Y M. Lossless watermarking algorithm for medical image's tamper detection and recovery with high quality[J]. Journal of Image and Graphics, 2014, 19(4): 583–591. [邓小鸿, 陈志刚, 毛伊敏. 基于无损水印的医学图像篡改检测和高质量恢复[J]. 中国图象图形学报, 2014, 19(4): 583–591. ] [DOI:10.11834/jig.20140413]
-
[10] Zhang Z W, Wu L F, Gao S B, et al. Robust reversible watermarking algorithm based on RIWT and compressed sensing[J]. Arabian Journal for Science and Engineering, 2018, 43(2): 979–992. [DOI:10.1007/s13369-017-2898-z]
-
[11] Bekkouch S, Faraoun K M. Robust and reversible image watermarking scheme using combined DCT-DWT-SVD transforms[J]. Journal of Information Processing System, 2015, 11(3): 406–420. [DOI:10.3745/JIPS.0200021]
-
[12] An L L, Gao X B, Li X L, et al. Robust reversible watermarking via clustering and enhanced pixel-wise masking[J]. IEEE Transactions on Image Processing, 2012, 21(8): 3598–3611. [DOI:10.1109/TIP.2012.2191564]
-
[13] Sui M, Li J B. Robust watermarking for medical images based on Arnold scrambling and DCT[J]. Application Research of Computers, 2013, 30(8): 2552–2556. [隋淼, 李京兵. 一种基于Arnold置乱变换和DCT的医学图像鲁棒水印算法[J]. 计算机应用研究, 2013, 30(8): 2552–2556. ] [DOI:10.3969/j.issn.1001-3695.2013.08.079]
-
[14] Yang S G, Li C X, Sun F, et al. Study on the method of image non-watermark in DWT domain[J]. Journal of Image and Graphics, 2018, 8(6): 664–669. [杨树国, 李春霞, 孙枫, 等. 小波域内图象零水印技术的研究[J]. 中国图象图形学报, 2018, 8(6): 664–669. ] [DOI:10.11834/jig.200306226]
-
[15] Xiao Z J, Jiang D, Zhang H, et al. Adaptive zero-watermarking algorithm based on boost normed singular value decomposition[J]. Journal of Image and Graphics, 2019, 24(1): 1–12. [肖振久, 姜东, 张晗, 等. 增强奇异值分解的自适应零水印[J]. 中国图象图形学报, 2019, 24(1): 1–12. ] [DOI:10.11834/jig.180443]
-
[16] Hahn E L. Nuclear induction due to free larmor precession[J]. Physical Review, 1950, 77(2): 297–298. [DOI:10.1103/PhysRev.77.297.2]
-
[17] Stejskal E O, Tanner J E. Spin diffusion measurements: Spin echoes in the presence of a time-dependent field gradient[J]. The Journal of Chemical Physics, 1965, 42(1): 288–292. [DOI:10.1063/1.1695690]
-
[18] Le Bihan D, Mangin J F, Poupon C, et al. Diffusion tensor imaging: concepts and applications[J]. Journal of Magnetic Resonance Imaging, 2001, 13(4): 534–546. [DOI:10.1002/jmri.1076]
-
[19] Kingsley P B. Introduction to diffusion tensor imaging mathematics: part Ⅲ. Tensor calculation, noise, simulations, and optimization[J]. Concepts in Magnetic Resonance Part A, 2006, 28A(2): 155–179. [DOI:10.1002/cmr.a.20050]
-
[20] Ling H F, Wang L Y, Zou F H. Real-time video watermarking scheme resistant to geometric distortions[J]. Journal of Electronic Imaging, 2011, 20(1): #013025. [DOI:10.1117/1.3565193]