A DWT-Based Robust Semi-Blind Image Watermarking Algorithm Using Two Bands

合集下载

毕业设计(论文)数字图像水印技术的研究与实现

湖南涉外经济学院毕业设计（论文）题目DWT域数字图像水印技术的研究与实现作者学部电气与信息工程学部专业通信工程学号指导教师黄彩云二〇一一年五月十日湖南涉外经济学院毕业设计（论文）任务书电气与信息工程学部通信工程系系（教研室）主任:（签名） 2010 年 12 月 18 日学生姓名: 学号: 专业: 通信工程1 设计（论文）题目及专题： DWT域数字图像水印技术的研究与实现2 学生设计（论文）时间：自 2011 年 1 月 8 日开始至 2011 年 4 月 25 日止3 设计（论文）所用资源和参考资料：[1] 陈武凡.小波分析及其在图像处理中的应用[J].科学出版社,2002, [2] 何东健.数字图像处理[J].西安电子科技大学出版社,2003，[3] 陈书海，傅录祥.实用数字图像处理[J].科学出版社,2005. [4] 陈桂明.应用MATLAB语言处理数字信号与数字图像[J].北京科学出版社,2000. [5] 汪小帆,戴跃伟,茅耀斌.信息隐藏技术方法与应用[J].北京机械工业出版社,2001.4 设计（论文）应完成的主要内容：就对目前数字水印技术的发展状况，包括数字水印的基本特征及分类，数字水印处理系统的基本框架以及目前的一些主要算法进行了论述。

最后围绕数字水印的两个最重要的特点——隐蔽性和鲁棒性进行考虑，设计并实现了一个完整的水印系统。

5 提交设计（论文）形式（设计说明与图纸或论文等）及要求：(1) 撰写设计报告；(2) 设计报告要求字数达2万字，提供电子版和文字版；(3) 设计报告包括目录、中英文摘要、关键词、方案选择及确定、技术要求、设计过程及参数计算、软件流程图及源程序、调试方法及步骤、小结等；(4) 提供电路原理图，要求用A0或A1图纸描绘。

6 发题时间： 2010 年 12 月 18 日指导教师：（签名）学生：（签名）湖南涉外经济学院毕业设计（论文）指导人评语[主要对学生毕业设计（论文）的工作态度，研究内容与方法，工作量，文献应用，创新性，实用性，科学性，文本（图纸）规范程度，存在的不足等进行综合评价]指导人：（签名）年月日指导人评定成绩：毕业设计（论文）评阅人评语[主要对学生毕业设计（论文）的文本格式、图纸规范程度，工作量，研究内容与方法，实用性与科学性，结论和存在的不足等进行综合评价]评阅人：（签名）年月日评阅人评定成绩：毕业设计（论文）答辩记录日期：学生：江堃学号： 200703402205 班级：通信工程0702 题目：DWT域数字图像水印技术的研究与实现提交毕业设计（论文）答辩委员会下列材料：1 设计（论文）说明书共页2 设计（论文）图纸共页3 指导人、评阅人评语共页毕业设计（论文）答辩委员会评语：[主要对学生毕业设计（论文）的研究思路，设计（论文）质量，文本图纸规范程度和对设计（论文）的介绍，回答问题情况等进行综合评价]答辩委员会主任：（签名）委员：（签名）（签名）（签名）（签名）答辩成绩：总评成绩：摘要随着计算网络和多媒体技术的快速发展，特别是Internet的普及，信息安全问题日益突出。

Robust Adaptive Watermarking...(IJIGSP-V7-N2-7)

I.J. Image, Graphics and Signal Processing, 2015, 2, 48-55Published Online January 2015 in MECS (/)DOI: 10.5815/ijigsp.2015.02.07Robust Adaptive Watermarking Based on Image Contents Using Wavelet TechniqueA. K. VermaElectrical and Electronics Engineering Department Hindustan Institute of Technology and Management, Agra, UP,IndiaEmail: ajaykrverma@C. PatvardhanElectrical Engineering Department Dayalbagh Educational Institute, Agra, UP, IndiaEmail: cpatvardhan@C. Vasantha LakshmiPhysics and Computer Science Department Dayalbagh Educational Institute, Agra, UP, IndiaEmail: cvasantha@Abstract—A good watermarking scheme should be able to perform equally well on all types of images irrespective of image contents because practically watermarking has to be applied to images of all types. In this paper, it is shown that in wavelet based spread spectrum technique, watermarking at level 1 decomposition is better for textured images while watermarking at level 2 decomposition is better for non-textured images to achieve maximum robustness against various types of attacks. The proposed wavelet decomposition level selection algorithm utilizes the edge histogram to classify the host image as textured or non-textured image and automatically selects the level of decomposition for robust watermarking. The use of Spread Spectrum watermarking technique and Bior6.8 wavelet, results better robustness. Performance of the proposed scheme and its relative effectiveness is demonstrated on both categories of images under different attacks.Index Terms—Wavelets, Bior6.8, Robustness, Correlation, Edge Histogram, Level of decomposition, Image Contents.I.I NTRODUCTIONThe rapid advancement of internet and computer technology has facilitated authorized as well as unauthorized manipulation and reproduction of digital multimedia contents. Therefore, design and development of effective digital multimedia copyright protection methods to prevent unauthorized duplication or tampering have become necessary in present time. To this end, watermarking of digital content has evolved as a possible solution. An efficient and effective watermarking technique has to satisfy at-least three major requirements i.e. Imperceptibility, Robustness and sufficient Payload amount. Simultaneously maximizing all the three together is difficult as they are non-commensurable. Therefore, finding a solution that provides satisfactory values of these parameters is a challenging task. Several watermarking algorithms are reported in the literature. Initial efforts were in the spatial domains [1, 2, 3]. Later, several improvements of frequency domain watermarking [4, 5, 6] were reported. Among frequency domain techniques, wavelet based watermarking schemes are more attractive [7, 8, 9, 10, 11] due to several advantages such as space-frequency localization and multi-resolution offered by the wavelet transform. Image contents play a very important role in deciding the performance of the watermarking process. An image may have several regions having different types of contents such as smooth (low frequencies) or textured (high frequencies). Smooth areas have low distortion resistance while textured areas can bear higher level of distortion. In the literature, several techniques are proposed which utilize image contents for effective watermarking. A DCT based watermarking method based on image contents is given in [12]. This method proposes the creation of a Just Noticeable Distortion (JND) mask which contains the JND values of each pixel. The mask is prepared on the basis of some image features such as Texture, Edges and Luminance satisfying Human Visual Systems (HVS). This method uses spread spectrum technique but results regarding quality of watermarked images and watermark extraction under various attacks are not shown. A DFT based watermarking scheme based on image contents is presented in [13]. In this scheme host image is divided into 9 sub images in 3×3. Watermark is embedded only in those sub images which are highly textured to avoid noticeable artifacts. The textured image blocks are identified by the use of Harris Corner detector. The PSNR achieved is around 40dB. The quality of recovered watermark is not shown. Only the presence and absence of watermark is highlighted. A DCT based watermarking method is given in [14]. This method prepares a mask based on image features such astexture, luminance, corners and edges. It embeds the watermark in those areas, which are less sensitive to human eyes. It uses decimal sequences of watermarking instead of random sequence. The PSNR achieved is in range of 35 to 38dB and quality of recovered watermark is not mentioned. Another method of watermarking based on local image features is proposed in [15]. This method is based on computation of a Noise Visibility Function (NVF) that characterizes the local image properties with high texture and edge regions, where watermark can be embedded strongly without resulting visible artifacts. This method achieves PSNR in range of 25 to 34 dB while quality of recovered watermark is not discussed. A wavelet based watermarking method based on image contents such as texture and luminance is given in [16]. In this approach, masking is done pixel by pixel based on texture and luminance. PSNR achieved in this method is around 35dB for various images. Watermark detection is shown instead of watermark recovery. A watermarking algorithm based on image contents in Ridgelet domain is proposed in [17]. In this method, image is first partitioned into small pieces. These pieces are classified as weak texture or strong texture according to the statistical properties of columns coefficients in Ridgelet Transform (RT). The middle frequencies of RT sub-bands are selected and watermarks are embedded in the higher energy directions of the pieces with strong texture which are less sensitively to human‘s vision. Another content based watermarking in Fourier domain is given in [18]. In this scheme, a perceptual mask is generated, which identifies both textured as well as smooth areas of the images. The embedding strength is then adjusted according to the embedding areas. The weighted PSNR is computed as quality of embedding. A hybrid image watermarking scheme based on DWT and SVD is proposed in [19]. In this approach, the edge information of an image is used to embed the watermark. In addition, the particle swarm optimization algorithm is used to search the proper value of watermark embedding strength. Experimental work shows the robustness of proposed scheme against various image processing attacks. Another watermarking algorithm robust against geometric attacks, based on image features is proposed in [20]. In this scheme, Watermark is embedded in the areas represented by the salient features of the image. It is shown that in an image, Salient features are resistive to the geometrical attacks. These salient features also provide reference points used in watermark embedding and detection.In all these methods proposed, the emphasis is on finding a suitable mask and adjusting the embedding strength of watermark accordingly. Most of the work is reported for the non-wavelet based techniques. Even in the case of few wavelet based schemes, the proposed content based methods do not explore the role and suitability of particular wavelet function, level of decomposition etc. in watermarking. Practically, whole image may belong to either smooth or textured category. How these proposed methods would deal with such cases, is not adequately discussed.In this paper an attempt has been made to identify the image type first on the basis of its contents and then applying suitable wavelet based watermark embedding scheme. The rest of the paper is organized as follows. Section 2 provides motivation for the work. Section 3 describes the wavelet decomposition level selection algorithm. Section 4 explains the proposed watermark embedding and watermark extraction algorithms. Section 5 shows the experimental results and some conclusions are given in section 6 followed by the references.II.M OTIVATIONIt is well understood that textured images have more high frequency components at a given level of decomposition. Some higher order wavelets such as Db10, Bior6.8 etc. capture these high frequency details and produce larger wavelet coefficients as compared to smooth images. Due to large wavelet coefficients the embedding strength is not enough to maintain the impact of watermarking, which results poor watermark recovery. This is illustrated by using eight test images which are shown in figure 1. Four images of top row are smooth (non-textured) and rest four are textured. The textured images are taken from standard texture image database of Brodatz [21].Lena Peppers BoatCameramanD15 D20 D84 D110Fig 1. Textured and Non-Textured Test ImagesThe effect of image contents on robustness for textured and non-textured images is shown in table 1. The spread spectrum watermarkingis done on these images with embedding strength and Bior6.8 wavelet. The watermark is embedded in diagonal wavelet coefficients (cD). The watermark embedding and extraction algorithms used in this process are explained in detail in section 4 of this paper. The energies of diagonal wavelet coefficients (embedding plane) at level-1 (EcD1) and at level-2 (EcD2) are also shown in table 1.The energies of diagonal coefficients show the amount of texture or randomness present in the image. Although sum of energies of all the detailed coefficients can also be considered for this purpose. The values of peak signal to noise ratio (PSNR) computed between original and watermarked images are shown in table 1. To show the quality of recovered watermark, Normalized Correlation Coefficient (NC) is computed between original and recovered watermark.Table 1. Effect of Image contents on robustnessImagesLevel 1 Level 2 Energy of cDsPSNR NC PSNR NC EcD1 EcD2 Lena 30.71 1.00 37.53 1.00 6.73 109.33 Peppers 30.71 1.00 36.98 1.00 91.44 94.66 Boat 30.64 1.00 37.33 1.00 69.75 119.66 Cameraman 31.39 1.00 38.08 1.00 138.83 146.97 D15 31.20 1.00 37.99 0.94 107.44 1372.80 D20 31.37 1.00 37.98 0.98 116.14 660.53 D84 31.70 1.00 38.27 0.96 116.32 1279.70 D110 31.64 1.00 38.15 0.76 275.73 3167.70 It can be clearly observed from table 1, that large wavelet coefficients (higher coefficients energies) are obtained for textured images as compared to smooth images which dilute the embedding. Therefore, poor robustness as indicated by reduced values of NC at level 2, is obtained for textured images. There may be two approaches by which the robustness of watermarking can be maintained for textured images at par with smooth images.Approach 1: The level of robustness can be increased by increasing the embedding strength for textured images while maintaining the level of decomposition (). Approach 2: An alternative approach is to reduce the level of decomposition for textured images for watermarking while maintaining embedding strength ().Both the approaches can maintain the desired level of robustness. Approach 2 is slightly better in terms of PSNR obtained. This can be seen as follows.In approach 1, required higher embedding strength reduces the PSNR. While in approach 2, modifications in major frequency components at level-1 also reduce the PSNR. If the value of PSNR is maintained equal in both the approaches with adjustment of, the value of NC will be slightly poorer in case of approach 1. At level-2 decomposition, the size of watermark embedding plane is smaller. This smaller size of coefficient matrix creates poor NC values as correlation is also affected by the size of data matrices. This slight reduction in NC of approach 1, can be overcome by further increase in embedding strength. But this will further reduce the PSNR. Therefore, to achieve same level of robustness, PSNR of approach 1 is slightly lower than that of approach 2.Therefore, approach 2 is adopted for textured images. The degradation in PSNR for textured images is not a major issue. Due to large high frequency contents available in textured images, human eyes are not able to discern this degradation.The discussion above concludes that non-textured images should be watermarked at level 2 while textured images should be watermarked at level 1 to achieve desired level of robustness for both types of images. This motivates a system which can identify the type of images (textured or non-textured) and watermark it accordingly. The level selection algorithm based on edge histogram is proposed in next section.III.L EVEL SELECTION ALGORITHMIn this section, approach 2 of previous section is described to maintain the desired level of robustness. The proposed level selection algorithm identifies the image type as to whether it is textured or non-textured and then applies either level-1 or level-2 watermarking accordingly. The scheme is outlined in figure 2.To classify the images based on the contents, the Edge Histogram Descriptor (EHD) of Mpeg-7 [22] is suitably adapted. For classification, the whole image is divided into 16 non-overlapping blocks in and each image block is then further divided into pixel blocks as shown in figure 3.From these sub blocks, the local edge orientation is captured for the following cases: vertical, horizontal, diagonal45, diagonal135 and isotropic (non-orientation specific) types of orientations. These are shown in table 2 along with their detector operators.Fig 2. Adaptive Watermarking Algorithm based on image contentsFig 3. Image division for EHD calculationThe histogram for each sub-image represents the relative frequency of occurrence of the 5 types of edges in the corresponding sub-image. As a result, each local histogram contains 5 bins (Vertical, Horizontal, Diagonal45, Diagonal135 and Isotropic). Since there are 16 sub-images in the image, a total ofhistogram bins are required to represent the imagecontents.Table 2. Five types of local edge orientations and their detectorsEdge TypeVisual RepresentationOperator MaskVertical Edge[]Horizontal Edge[] Diagonal[√ √] Diagonal[ √ √] Non-Orientation TypeEdges[]The edge orientation of a sub block is captured by applying all five detectors as follows.∑ (1)Where,[ ] []represents image subblock and [ ] [] represents the edge detector.For one image block, by applying all these five operators on a image sub block, the five values are obtained and maximum of these five is compared with a threshold value ( ) to find the dominant edge orientation as, = .The count of corresponding bin is increased by one and it is repeated for all sub blocks. Thus a first image block out of total 16 is represented by 5 bins as . This process is repeated for all remaining 15 image blocks getting their 5 bins representation. For example, 2nd block representation is and last block representation is . From this five value representation of all 16 image blocks, a matrix is prepared as shown in (2).[ ](2)The value of is computed by taking column wise and row wise mean of matrix as,( ) (3)In the proposed scheme, the value of is taken as decision level for finding whether an image is textured or non-textured. To validate the results of image classification algorithm based on image contents, several textured and non-textured images are taken for experiment. For this purpose, total 107 images of size of both the categories are collected from various sources including standard image databases and Word Wide Web. The results of image separation on the basis of their are shown in table 3 and in figure 4.In figure 4, the mean values of is also shown for both the categories of the images. Results shown in table 3 and figure 4 show the classifications of selected images on the basis of their edge histograms. There may be the possibility that some images lie on the verge of classification. This condition can lead to miss-classification of the images. But this is not the problem because in this scheme, the objective is not to implement the precise classification of images. The simple edge histogram based algorithm is sufficient to facilitate the proper watermarking of smooth and textured images to achieve better robustness.Table 3. Results of Image classification based on image contents Image Type No. of ImagesMin. Max. Avg.Smooth(Low Frequency) 107 0.00 87.09 14.19 Textured(High Frequency)107122.18643.15300.59Fig 4. (a) profile of smooth images, (b)profile of textured imagesIV. W ATERMARKING S CHEMEThe spread spectrum watermarking [6] is known to be robust against various types of attacks. In this paper, spread spectrum watermarking is used for watermark embedding and extraction for both types of images. Watermark embedding and extraction algorithms are explained in next subsections. A. Watermark Embedding AlgorithmThe steps of the proposed embedding algorithm are as follows.Input : A grayscale image of type uint8 and of size and a binary watermark.Output: Watermarked Image.1.The input image is categorized into either texturedor smooth image by the level selection algorithm described. If image is textured then decomposition level otherwise for smooth images.2.Perform N-level wavelet decomposition of inputimage to obtain four coefficients matricesof size.3.Select high frequency diagonal wavelet coefficientmatrix for watermark embedding.4.Select a seed to generate a Pseudo Random Sequenceof size equal to the size of frequency band and modify it to obtain matrix using the relation,,.5.If watermark bit is 0 (Black) then modify thewavelet coefficients as, where, isembedding strength. If watermark bit is 1 (white) then wavelet coefficients are left unchanged.6.Repeat the step 4 and 5 for all ‗0‘ watermark bits witha newly generated sequence for each bit.7.Take modified to its original position and takeinverse DWT to get watermarked image.pute the PSNR for and to determine thechange in host image.The embedding algorithm is shown in figure 5.Fig 5. Watermark Embedding Algorithm for content basedwatermarkingB. Watermark Extraction AlgorithmFollowings are the steps of watermark extraction, Input: Watermarked Image and Seed value. Output: Extracted Watermark.1.Decompose the watermarked image in levelsand obtain the modified coefficients matrix.2.Generate the same random sequence, whichwas generated in embedding process using same seed value and convert the random sequence intowith.pute the correlation coefficients between randomsequence and the modified coefficient matrixas Repeat step 2 and 3 for all watermark bits and compute all correlation values.pute the mean correlation value ( )and initialize a row matrix having all values ‗1‘ equivalent to the size of the watermark.5.For every watermark bit, compare with andmodify the as follows,{6.Reshape the row matrix into a matrix of size oforiginal watermark matrix to get recovered watermark.pute the NC between original watermark (W) andrecovered watermark to observe the quality of extracted watermark.The watermark extraction algorithm is shown below graphically in figure 6.Fig 6. Watermark Extraction Algorithm for content based watermarking V.E XPERIMENTAL RESULTS AND ANALYSISIn this section, the proposed methodology is tested on various textured and non-textured images. These test images are already shown in figure 1. All test images are grayscale and of size. The binary watermark used is shown in figure 7.Fig 7. Binary Watermark ()While computing of equation 3, the threshold value is taken as 100. For the purpose of automatic wavelet decomposition level selection, the value of is chosen to be 120 as decision level.This decision level will categorize the input host images as textured and non-textured images. The input images, those having below 120 will be categorized as smooth images, while those havingabove 120 will be considered as texturedimages. The decision level 120 is derived empirically by computing with amount of texture variation in input host image as shown in table 4.Table 4. Watermarking evaluation with texture variationImagesLevel 1 Level 2 PSNR NC PSNR NCImage:, Texture:(0.00%)30.48 1.0037.04 1.0029.53(1.53%)31.71 1.00 38.20 1.00 34.05(6.25%)31.76 1.00 38.47 1.00 47.18(14.06%)31.79 1.00 38.47 1.00 69.00(25.00%)31.90 1.00 38.47 1.00 100.18(39.06%)31.90 1.00 38.54 0.9832 138.61 Texture: (56.25%)31.90 1.00 38.54 0.9750 185.94 Texture: (76.56%)31.90 1.00 38.54 0.9589 241.16 Texture: (100.00%)32.09 1.00 38.54 0.8530 302.68 All images in table 4 are watermarked with using both level-1 and level-2 decomposition with Bior6.8 wavelet. This wavelet is chosen because Bior6.8 is a symmetric, biorthogonal wavelet with linear phase and gives better reconstruction with minimum error. The Peak Signal to Noise Ratio (PSNR) is calculated in dB between the watermarked image and original image. Normalized Correlation Coefficient (NC) between recovered watermark and original watermark is calculated under no attack condition.From table 4, it can be observed that when the value ofis 138.61 (39.06% texture), watermark recovery in level 2 is degraded with NC lesser than 1. Therefore, it is better to choose level 1 watermarking to improve watermark extraction.Similarly when the value of is 100.18, level-2 watermarking performs well in watermark extraction. Thus, an intermediate value of 120 is chosen as decision level for automatic level selection for textured and non-textured images aiming at better watermark extraction. The analysis shown in table 4 is done for image size. For an image half of its size the decision level can be down scaled by 0.5 and similarly for image double of its size, it can be up scaled by 2 assuming that the amount of texture is proportional to the image size. It can be seen from table 4 that although decomposition at level 1 improves watermark recovery for textured images, the value of PSNR reduces. This is not a big problem as already discussed because due to more texture available in the image, this small degradation in image is almost imperceptible to the human eyes.The performance of watermarking scheme along with level selection algorithm is tested under various attacks such as JPEG compression, Cropping, Noise addition, Filtering etc. and the results are shown in the tables 5 to 11.The shaded areas in tables 5 to 11 show the preferred level watermarking for smooth and textured images depending on the value of. The value of NC calculated between original and recovered watermark clearly shows that in all attacks the robustness is better for smooth images if they are watermarked at level 2 while it is better for textured images, if they are watermarked at level 1. The only exception is the case of blurring attack, where level 2 watermarking is more robust for textured images. This happens because, blurring attack removes high frequency details of the textured images, therefore, making them in the category of smooth images.Table 5. Comparison of NC under JPEG Compression (Q=30)ImageLevel1 Level2PSNR NC PSNR NCTable 6. Comparison of NC under Salt & Pepper Attack (Strength=0.1) ImageLevel1 Level2Table 7. Comparison of NC under Gaussian Noise AttackImageLevel1 Level2Table 8. Comparison of NC under Sharpening Attack(Mask= [-1 -1 -1;-1 9 -1;-1 -1 -1])ImageLevel1 Level2Table 9. Comparison of NC under Median Filtering Attack(Mask)ImageLevel1 Level2 PSNR NC PSNR NCLena 8.20 30.71 0.9052 37.53 0.9833 Pepper 29.53 30.71 0.7760 36.98 0.9916 Boat 30.64 30.64 0.8007 37.33 0.9916 Camera Man 84.74 31.39 0.7760 38.08 1.0000 D15 363.26 31.20 0.9510 37.99 0.7430 D20 260.71 31.37 0.8979 37.98 0.7324 D84 364.61 31.70 0.8172 38.27 0.8221 D110 402.08 31.64 0.6411 38.15 0.6182 Table 10. Comparison of NC under Blurring Attack( Averaging Mask)ImageLevel1 Level2 PSNR NC PSNR NCLena 8.20 30.71 0.6712 37.53 1.0000 Pepper 29.53 30.71 0.5994 36.98 1.0000 Boat 30.64 30.64 0.5398 37.33 1.0000 Camera Man 84.74 31.39 0.7004 38.08 1.0000 D15 363.26 31.20 0.6548 37.99 0.8741 D20 260.71 31.37 0.6258 37.98 0.9277 D84 364.61 31.70 0.5600 38.27 0.9201 D110 402.08 31.64 0.5046 38.15 0.6969Table 11. Comparison of NC under Cropping Attack(1/4th upper left is cut)ImageLevel1 Level2VI.C ONCLUSIONSIn this paper, an adaptive robust spread spectrum method of digital image watermarking in wavelet domain is proposed. The algorithm automatically selects the level of wavelet decomposition based on image contents to provide maximum robustness to both textured and non-textured images. The proposed scheme is tested for various types of images. The results show that there is good improvement in the quality (Correlation Coefficients) of recovered watermark under various attacks for textured images if they are watermarked at level 1. Though the watermarking at level 1 reduces the PSNR value of textured images slightly but better robustness is achieved. The small degradation in the value of PSNR for textured images is not a problem as due to more intensity variations in textured image, this small degradation of image quality is almost insensitive to human eyes. Similarly for smooth images, level 2 decomposition works well, where better robustness with high PSNR is achieved. High PSNR is necessary in case of smooth images to avoid any noticeable visual artefacts in the image.R EFERENCES[1]R. G. Van Schyndel, A. Z. Tirkel, C. F. Osborne, ―Adigital watermark‖, IEEE Proceedings ICIP, vol.2, pp. 86-90, 1994.[2]N. Nikolaidis, I. Pitas, ―Copyright protection of imagesusing robust digital signatures‖, IEEE International Conference on Acoustics, Speech Signal Processing, vol.4, pp.2168-2171, May 1996.[3]R. Wolfgang, E. Delp, ―A watermark for digital image‖,IEEE Proceedings ICIP, Vol.3, pp.211-214, 1996.[4]Emir Ganic, Scott D. Dexter, Ahmet M. Eskicioglu,―Embedding Multiple Watermarks in the DFT Domain using Low and High Frequency Bands‖, Proceedings on SPIE international conference on Security, Steganography, and Watermarking of Multimedia ContentsVII, Vol. 5681, 2005.[5]Lin S. D., Chin Feng Chen, ―A robust DCT-basedwatermarking for copyright protection‖, IEEE Transactions on Consumer Electronics, Vol. 46, Issue.03, pp. 415-421, August, 2000.[6]I. J. Cox, Joe Kilian, F. Thomson Leighton, TalalShamoon, "Secure Spread Spectrum Watermarking for Multimedia", IEEE Transactions on Image Processing, Vol. 6, No. 12, December 1997.[7] C. V. Serdean, M. Tomlinson, J. G. Wade, M. A.Ambroze, ―Protecting Intellectual Rights: Digital WM in the Wavelet Domain‖, Proceedings of the IEEE International Workshop on Trends and Recent Achievements in Information Technology, 16-18 May 2002.[8] C. Patvardhan, A. K. Verma, C. Vasantha Lakshmi, "AComparative Analysis of Spread Spectrum Watermarking Technique in Wavelet Domain", Journal of Computer Science and Engineering (JCSE), U. K., Vol. 9, Issue 2, pp. 38-43, October 2011.[9] C. Patvardhan, A. K. Verma, C. Vasantha Lakshmi, "ARobust Wavelet Packet Based Blind Digital Image Watermarking using HVS characteristics", International Journal of Computer Applications (IJCA), Vol. 36(9), pp. 06-12, December 2011.[10] Nikita Kashyap, G. R. Sinha, "Image Watermarking Using3-Level Discrete Wavelet Transform (DWT)", International Journal of Modern Education and Computer Science, MECS, Vol. 3, pp. 50-56, 2012.[11] M. Natarajan, Y. Govindarajan, "PerformanceComparison of single and multiple watermarking techniques", International Journal of Computer Network and Information Security, MECS, Vol. 7, pp. 28-34, 2014. [12] M. S. Kankanhalli, Rajmohan, K. R. Ramakrishnan,"Content Based Watermarking of Images", Proceedings of the 6th ACM International Conference on Multimedia, pp. 61-70, Bristol, UK, September 1998.[13] Xiaojun Qi, Ji Qi, "A robust content-based digital imagewatermarking scheme", Elsevier Journal of Signal Processing, Vol. 87, pp. 1264-1280, 2007.[14] A. K. Parthasarathy, SubhashKak, "An Improved Methodof Content Based Image Watermarking", IEEE Transactions on Broadcasting, Vol. 53, No. 02, pp. 468-479, 2007.[15] Sviatoslav Voloshynovskiy, Alexander Herrigel, NazaninBaumgaertner, Thierry Pun, "A Stochastic Approach to Content Adaptive Digital Image Watermarking", Proceedings of 3rd Springer International Workshop on Information Hiding (IH), pp. 211-236, 1999.[16] Mauro Barni, Franco Bartolini, Alessandro Piva,"Improved Wavelet-Based Watermarking Through Pixel-Wise Masking", IEEE Transactions on Image Processing, Vol. 10, No. 05, pp. 783-791, 2001.[17] Zhihua Xie, Shengqian Wang, Lixin Gan, Lin Zhang,Zhenghua Shu, "Content Based Image Watermarking in the Ridgelet Domain", IEEE International Symposium on Electronic Commerce and Security (ISECS), pp.877-881, 2008.[18] F. Autrusseau, P. Le Callet, A. Ninassi, "A study ofcontent based watermarking using an advanced HVS model", IEEE International Conference on Intelligent Information Hiding and Multimedia Signal Processing, pp. 485-488, 2007.[19] Chih-Chin Lai, Chi-Feng Chan, Chen-Sen Ouyang, Hui-Fen Chiang, "A Robust Feature-Based Image Watermarking Scheme ", Proceedings of 4th IEEE International Conference on Software Engineering, Artificial Intelligence, Networking and Parallel/Distributed Computing (SNPD), pp. 581-585, July 2013.[20] Chun-Hung Chena, Yuan-Liang Tangb, Chih-PengWangb, Wen-Shyong Hsieha, "A robust watermarking algorithm based on salient image features", Optik - International Journal for Light and Electron Optics, Elsevier, Volume 125, Issue 3, pp. 1134–1140, Feb. 2014.[21] Brodatz Texture Database, Available onhttp://www.ux.uis.no/~tranden/brodatz.html.[22] Chee Sun Won, Dong Kwon Park, and Soo-Jun Park,―Efficient U se of MPEG-7 Edge Histogram Descriptor‖, ETRI Journal, Volume 24, Number 1, February 2002.Authors ’ ProfilesDr. A. K. Verma is working in Hindustan Institute of Technology and Management, Agra, UP, India as Associate Professor in the department of Electrical and Electronics Engineering. He obtained his B.Sc.-Engg. in Electrical Engineering in year 2000 and M.Tech. in Engg. Systems in year 2002 from Dayalbagh EducationalInstitute (DEI). He was awarded with Gold Medals in both B.Sc.-Engg. and M.Tech. for securing highest marks. He has obtained his Ph.D. in March 2014 from DEI. His areas of research is wavelets and their applications in Signals and Image Processing. He has published more than 25 research papers in various international journals and conferences of repute. He is also reviewer of few international conferences. Dr. Verma is life member of ISTE, India.Prof. C. Patvardhan is working in the Dayalbagh Educational Institute (DEI), Agra, UP, India as Professor in Electrical Engineering department. He obtained his B.Sc. (Engg.) from DEI in 1987, M.Tech. from IISc, Bangalore in Computer Science in 1989 and his Ph.D. in 1994. He has published more than 250 papers in Journalsand Proceedings of Conferences and has won 20 Best Paper Awards. He has also published one book and has been an Editor of two Conference proceedings. He has been the PI and Co-PI of several funded R&D projects. His current research interests are Quantum and Soft Computing, Image Processing and he is a reviewer for International Journals and Conferences. He is a life member of Computer Society of India, Systems Society of India and Indian Science Congress Association and a Fellow of United Writers Association of India, Chennai.Dr. C. Vasantha Lakshmi is working in the Dayalbagh Educational Institute, Agra, UP, India as an Associate Professor in the Department of Physics and Computer Science. She obtained her B.Tech. (ECE) from JNTU in 1992, M.Tech. (CS) from Central University, Hyderabad in 1994 and Ph.D. from DEI, Dayalbagh in 2003. Shehas been the PI and Co-PI of several funded research projects. She has authored one book and has more than 40 publications in Journals and Conferences. Her research has been recognized by several awards including the Indian Science Congress Young Scientist Award in 2004 and Systems Society of India Young System Scientist Award in 2009. Her research focuses on Image Processing and Pattern Recognition. She is a member of the IEEE and the IEEE Computer Society.Manuscript received Month Date, Year; revised Month Date, Year; accepted Month Date, Year.。

DWT低频域自调整提取的盲水印算法

（ｔｎｆｒａｉｅｈｏｇｏｅｅＨｎｈｎＴａｈｒ’ ｏｌｅＣａｚｏ２０１，ｕｎｄｎ，ｈｎ）ＭａｈａｄＩｏｍｔｎＴｃｎｌｙＣｌｇ，ａｓａｅｃｅＣｌｇ，ｈｏｈｕ５１４ＧａｇｏｇＣｉａｎｏｏｌｓｅ。Ｆｃｌｏｐｔ，ｕｎｄｎｎｖｒｔｆＴｃｎｌｙＧａｇｈｕ５０９Ｇａｇｏｇ，ｈｎ）（ａｕｔｏｍｕｅＧａｇｏｇＵｉｓｙｏｅｈｏｇ，ｕｎｚｏ１００，ｕｎｄｎＣｉｙｆＣｒｅｉｏａ
ｎｗｌｄｗａｅａｋｎｌｏｔｍａｅｎｔｅｌｗｓｆｎｕｎｙｄｍａｎｏｉｃｅｅｗｖｌｔｒｎｆｒｓｐｏｏｅ．ｎｔｉｃｅ，ｘａｔｇｅｂｉｔｒｒｉｇａｇｒｈｂｓｄｏｈｏｅｔｒｑｅｃｏｉｆｓｒｔａｅｅａｓｏｉｒｐｓｄＩｈｓｓｈｍｅｅｔｃｉｎｍｉｅｄｔｍｒｎ
ｇｉｔｍｏｔｃｍｍｏｄｓｏｔｏ．ａｎｓｓｏｎｉｔｒｉｎｓ
Ｋｅｗｏｄｙｒｓ
ＤｓｒｔｖｌｔｒｓｏＢｌｄａｇｒｈｉｃｅｅｗａｅｅａｆｒｔｎｍｉｌｏｔｍＨｕｎｉａｙｔｍＥｔａｔｇｉｔｎｉｎｉｍａｖｓｌｓｓｅｕｘｒｃｉｎｅｓｔｎｙ
明版权。
频成分又是图像的平滑部分，微的改动也可能导致图像失真轻并引起视觉注意，算法难度较大。Ｘｅ算法就是一种基于ＤｉＷＴ低频域的盲水印算法。本文在Ｘｅ算法的基础上提出一种新的ＤｉＷＴ低频域盲水印算法，印信息是有意义的二值图像。该算法克服Ｘｅ算法水ｉ嵌入量少、嵌入强度固定的缺点，结合ＨＶＳ特性自适应嵌入，提高水印的鲁棒性。该算法最大的特点是：印提取时能根据嵌水入图受攻击的状态自动调整提取强度，以获取最佳的提取效果。

一种基于DWT的盲图像水印算法

一
种基于ＤＷＴ的盲图像水印算法
雷贲张专成２杨慧雅，，
（．１武警工程学院研究生队，陕西西安７０８；．１０６２武警工程学院通信工程系，陕西西安７０８）１０６
摘要：提出了一个新的基于量化抖动和模式分组编码的盲数字图像水印算法，该算法首先在ＤＴ离小波变换）Ｗ（散的第
ｔａｉｍｔｏａｏｄｐｒｒａｃｎ￥ｍｉｔａｅｐｏｅｓｇｈｔｓｅｈｈａｇｏｆｎｅｏｏｅｄｇａｉｇｒｃｓｎ．ｈｔｄｓｅｏｍｉｌｍｉ
Ｋｅｒｓｑａｔａｉｎ；ｇｏｐｃｄｎ；ｉｇｅｗｔｒａｋｎｙｗｏｄ：ｕｎｉｔｚｏｒｕｏｉｇｍａａｍｒｉｇ；ｐｔｍｅａｔｅ
维普资讯
２０年第５期０８
文章编号：０６２７（ｏ８０－０００１０－４５２ｏ）５０２－２
计算机与现代化ＪＳＡＪＹＩＮＡＨＵＩＵＮＩＵＸＡＤＩＡ
数抖动的范围不大，可满足不可见性的要求，是所修改的高频系数很难但抵抗低通滤波处理的攻击，具有很好的鲁棒性；不对
于低频系数来说系数绝对值较大，其修改后系数抖对动范围比较大，因此透明性较差，是对于大部分攻但
二层中频子带中随机选取相应位置的绝对值最大的系数作为水印的嵌入区间，小波系数进行量化并根据分组编码原对
理生成预嵌入模式，然后将一个二值水印图像嵌入到模式当中。实验证明通过量化抖动和模式分组编码相结合的方法，

基于HD和SVD的DWT变换的数字图像水印

基于HD和SVD的DWT变换的数字图像水印作者：甘志超刘丹来源：《现代信息科技》2022年第01期摘要：文章提出一种基于离散小波变换（DWT）、Hessenberg分解（HD）和奇异值分解（SVD）的图像水印方法。

在嵌入过程中，对原始载体图像进行多级DWT分解，并将得出的子带系数作为HD的输入。

在创建水印的同时对SVD进行操作，通过缩放因子将水印嵌入到主图像中。

运用果蝇优化算法，通过给出的客观评价函数来寻找比例因子。

在各种欺骗攻击下，将所提出的方法与其他方法进行比较，实验结果表明，该方法对水印具有良好的鲁棒性和不可见性。

关键词：图像水印;离散小波变换;Hessenberg分解;奇异值分解中图分类号：TP391.4 文献标识码：A文章编号：2096-4706（2022）01-0040-04Abstract： This paper proposes an image watermarking method based on discrete wavelet transform （DWT）， Hessenberg decomposition （HD） and singular value decomposition （SVD）. In the embedding process， the original carrier image is decomposed by multi-level DWT， and the obtained subband coefficients are used as the input of HD. While creating the watermark， the SVD is operated， and the watermark is embedded into the main image through the scaling factor. Using the fruit fly optimization algorithm， the scale factor is found through the given objective evaluation function. Under various spoofing attacks， the proposed method is compared with other methods. The experimental results show that this method has good robustness and invisibility to watermark.Keywords： image watermarking; discrete wavelet transform; Hessenberg decomposition; singular value decomposition0 引言图像的鲁棒性和不可见性是评价水印技术有效性的两个主要指标。

彩色图像的信息隐藏技术研究

彩色图像的信息隐藏技术研究
/article/2012-2-10/615-1.html
摘要:数字水印技术是解决数字作品版权保护问题的非常有效的方法,它在灰度图像中的应用已进行了广泛、深入的研究,而目前该技术主要是针对彩色图像和视频领域.先前众多的 DW T(discrete wavelet transform)域水印算法几乎都是私有水印或易脆水印,而针对版权保护的鲁棒的、公开的小波域水印算法却极少.在充分考虑人类视觉系统掩蔽特性的基础上,首先给出基于图像分块的临界噪声阈值矩阵 JND(just noticed difference),进而提出了一种基于 DW T 的鲁棒公开水印技术.算法首先把原图像各 8×8 块按 Hilbert 扫描顺序排列,然后在原图像分块的 Hilbert 序列中选取两相邻块分别进行一层 DWT,再结合各分块 JND 阈值,通过不同强度地调整两相邻块各对应细节子带均值之间的大小关系以自适应地嵌入水印.嵌入水印具有很好的透明性,水印嵌入强度是与原图像特征相自适应的.同时,水印的提取无须求助于原图像. 此外,实验结果也证明,该算法对常见图像处理操作、图像缩放和裁剪、挤压、扭曲等几何变换有较高的鲁棒性,特别是信号增强操作处理几乎不影响水印的正确提取,所以该算法是有效和实用的.
Hilbert scanning order. Then two neighboring blocks are selected from the Hilbert sequen ce of the host image blocks in turn, and 1-level DWT is applied to the two chosen block s. Finally, a corresponding detail subband is chose from three detail subbands of the two neighboring blocks at a time, respectively. A binary watermark with visually recognizable patterns is embedded into the host image by modifying the polarity of the average value of the two corresponding subbands. The embedded watermark is invisible to human eyes a nd adapted to the original image by exploiting the HVS masking characteristics. The expe rimental results show that the proposed algorithm is effective and robust to common imag e processing operations and some geometric distortions such as cropping, pinching, pixel-s hift, and so on, especially, it receives better robustness under signal enhancement operation s. So a conclusion can be made that the proposed technique is practical.

基于小波变换的用于医学图像的半脆弱水印算法

基于小波变换的用于医学图像的半脆弱水印算法蔡键;叶萍;刘涛【摘要】According to the speciality of medical images, this paper proposes a semi-fragile watermarking algorithm applied for medical images. The recovery watermark and authentication watermark are embedded into the Hypo-LSB and LSB of the primitive image respectively.The high 6 bits of the primitive image are conducted wavelet transform, and then the recovery watermark is obtained from the quantification and chaotic scrambling of the low-frequency wavelet coefficients and is embedded into the Hypo-LSB of the image. The small volume of information is embedded into LSB of the primitive image to generate authentication watermark. Experiments result shows that the proposed algorithm can locate accurately the position where the primitive image is tampered, and can also restore the region with tampered contents.The authentication watermark in this algorithm can authenticate the medical images effectively. So the algorithm can meet the special request of the medical image.%针对医学图像的特殊性,提出了一种应用于医学的半脆弱水印算法.原始图像的次低位和最低位分别嵌入恢复水印和认证水印.原始图像的高6位进行小波变换,低频系数经量化和混沌置乱后生成恢复水印,将其嵌入到图像的次低位.小容量信息嵌入到原始图像的最低位,生成认证水印.实验结果表明该算法可以准确定位原始图像被篡改的位置,并能对内容被篡改的区域进行恢复.算法中的认证水印能够对医学图像进行有效地认证.该算法能够较好地满足医学图像特殊性的要求.【期刊名称】《计算机应用与软件》【年(卷),期】2011(028)006【总页数】4页(P278-281)【关键词】半脆弱水印;图像认证;医学图像;恢复水印【作者】蔡键;叶萍;刘涛【作者单位】徐州师范大学现代教育技术中心,江苏徐州,221116;中南林业科技大学理学院,湖南长沙,410004;中南林业科技大学计算机与信息工程学院,湖南长沙,410004【正文语种】中文0 引言医学图像通常存储在PACS(Picture Archivingand Communication Systems)，可以通过网络进行存取，用于医疗诊断［1］。

Matlab的第三方工具箱全套整合(强烈推荐)

CompEcon- solving problems in economics and finance
Complex- for estimating temporal and spatial signal complexities
Computational Statistics
Coral- seismic waveform analysis
2007-8-2915:04
#1
littleboy
助理工程师
精华0
积分49
帖子76
水位164
技术分0
JMatLink- Matlab Java classes
Kalman- Bayesian Kalman filter
Kalman Filter- filtering, smoothing and parameter estimation (using EM) for linear dynamical systems
DMsuite- differentiation matrix suite
DMTTEQ- design and test time domain equalizer design methods
DrawFilt- drawing digitaland analog filters
DSFWAV- spline interpolation with Dean wave solutions
FSBOX- stepwise forward and backward selection of features using linear regression
GABLE- geometric algebra tutorial
GAOT- genetic algorithm optimization

基于DWT的图像内容半脆弱水印认证算法

（Institute of Karst Geology of Chinese Academy of Geological Sciences ， Guilin 541004 ， China） Abstract： Semi- fragile watermarking because of the important role in multimedia content authentication and close attention by people. In order to be able to distinguish between accidental attacks and malicious tampering, semi- fragile watermarking need to general image content protection operation has certain this algorithm firstly to a layer of DWT wavelet transform of the image, in the image transform high
收稿日期： 2016-07-19 作者简介：张劲松（ 1987），男，汉，在读硕士，助理研究员，研究方向:数据挖掘，网络编程。 E-mail： xiahuaxianyou@。
第6期
张劲松等 . 基于 DWT 的图像内容半脆弱水印认证算法
LL2 HL2 HL1 LH2 的中 2ˑ2 子块 HH1
摘
要：半脆弱水印因为在多媒体内容认证方面的重要作用而受到人们密切的关注。为了能够区分偶
然攻击与恶意篡改，半脆弱水印需要对一般的内容保护图像操作有一定的鲁棒性。本文提出了一种新的基于 DWT 变换的半脆弱水印算法，该算法思想首先对图像进行一层 DWT 小波变换，再把图像变换出的高频系数部分（LH1 和 HL1）进行分块，然后分别算出每小块的能量值，根据能量值大小关系嵌入水印。仿真实验结果表明，该算法特别在 JPEG 压缩方面有较好的鲁棒性，另外算法具有精确的定位检测能力，算法整体性能较好。关键词：半脆弱水印； DWT； JPEG；鲁棒性；小波变换中图分类号： TP301.6 文献标识码： A 文章编号： 1001-7119 （2017） 06-0192-04

基于超混沌和DWT变换的灰度图像盲水印算法研究

基于超混沌和DWT变换的灰度图像盲水印算法研究蒋爱平;王承琨;齐可心【摘要】数字图像水印技术为数字图像版权提供了有效的保护措施.利用超混沌将灰度图像水印进行编码和加密,同时在编码过程中加入了校验位,这与传统的水印置乱和编码方式相比具有更好的安全性和可恢复性.算法使用了小波变换,使嵌入水印后的图像具有一定的鲁棒性.该算法在提取水印的情况下不需要原始图片参与,实现了盲提取的效果.利用Matlab软件仿真得到了测试结果,从峰值信噪比和归一化互相关系数两个方面分析了该算法的性能,可知该算法具有良好的鲁棒性和隐藏性.【期刊名称】《黑龙江大学自然科学学报》【年(卷),期】2016(033)004【总页数】5页(P550-554)【关键词】灰度水印;超混沌;盲水印;小波变换;【作者】蒋爱平;王承琨;齐可心【作者单位】黑龙江大学电子工程学院,哈尔滨150080;黑龙江大学电子工程学院,哈尔滨150080;黑龙江大学电子工程学院,哈尔滨150080【正文语种】中文【中图分类】O436近几年,数字图像水印技术引起了学者们的广泛关注,相继有脆弱水印、盲水印、鲁棒水印等多种适用于不同场合的数字图像水印。

按照水印的实现方法，可将数字水印分为时域水印和频域水印，由于时域水印的鲁棒性差，很容易被修改或删除，人们把更多的目光放到了频域水印上。

数字图像的频域变换有离散傅里叶变换、离散余弦变换、离散小波变换、瘠薄变换等多种方式，将数字图像进行频域变换后能够抵抗一定的攻击，水印提取效果更加优秀。

为了保证水印信息的安全，一般情况下会将水印信息进行置乱与加密，常用的置乱方式为Arnold置乱，文献[1]中所涉及的就是传统的Arnold置乱方法，利用Arnold置乱的周期特点，将周期当作密匙。

虽然该置乱方式有较好的置乱效果，但利用穷举法很容易就可以将其破译。

文献[2]中详细的介绍了这几年对Arnold置乱的研究现状，虽能提高置乱效果，但运行效率低、加密效果不强，仍没有摆脱周期性的缺点。

基于顶点统计特征的三维网格模型盲水印算法

基于顶点统计特征的三维网格模型盲水印算法钱逸;王新宇;詹永照【期刊名称】《计算机辅助设计与图形学学报》【年(卷),期】2015(027)009【摘要】To obtain a better balance between the robustness and transparency of the blind watermarking scheme for three-dimensional mesh models, a novel blind watermarking scheme based on statistical charac-teristic of vertices is proposed in this paper. In the embedding process, the distances from vertices to the model center are calculated as their eigenvalues based on which vertices are separated into bins that are fur-ther classified into three groups: united bins, recovery bins and buffer bins according to the distribution of eigenvalues. Watermark is embedded by adjusting the distribution of normalized eigenvalues in united bins and then the position of the model center is recovered through modifying the coordinates of vertices in re-covery bins. The proposed scheme enhances the robustness owing to the comprehensive utilization of united bins, recovery bins and vertex fine-tuning method while achieves the blind detection of watermark. The ex-perimental results show that the proposed scheme not only has a good performance in transparency but also can effectively resist the common attacks including translation, rotation, uniform scaling, vertex rearrange-ment, noise, smoothing, subdivision and remeshing.%针对三维网格模型盲水印算法鲁棒性和透明性未很好权衡的问题, 提出一种基于顶点统计特征的三维网格盲水印算法. 在水印嵌入过程中, 首先计算顶点到模型中心的距离作为顶点的特征值; 根据该特征值对顶点进行分区, 并依据特征值的分布将分区分为联合嵌入分区、恢复分区和缓冲分区; 然后通过改变联合嵌入分区内归一化后的顶点特征值的分布来嵌入水印, 并修改恢复分区中顶点的坐标还原模型中心. 文中算法通过综合利用联合嵌入分区、恢复分区和顶点微调方法增强了水印的鲁棒性, 并实现了水印的盲检测. 实验结果表明, 在保持较好的透明性的基础上, 该算法能够有效抵抗平移、旋转、均匀缩放、顶点重排序、噪声、平滑、细分和网格重构等常见攻击.【总页数】8页(P1661-1668)【作者】钱逸;王新宇;詹永照【作者单位】江苏大学计算机科学与通信工程学院镇江 212013;江苏大学计算机科学与通信工程学院镇江 212013;江苏大学计算机科学与通信工程学院镇江212013【正文语种】中文【中图分类】TP391.41【相关文献】1.基于固定正交基频谱分析的三维网格模型盲水印算法 [J], 徐涛;张艳宁2.基于三维网格模型的双重数字盲水印算法 [J], 唐斌;康宝生;王国栋;康健超;赵建东3.基于顶点几何数据构造的三维网格模型零水印算法 [J], 徐涛;汪华斌;李慧;4.基于顶点几何数据构造的三维网格模型零水印算法* [J], 徐涛; 汪华斌; 李慧5.基于小波变换的三维网格模型盲水印算法 [J], 杨明鹏;叶帼华因版权原因，仅展示原文概要，查看原文内容请购买。

英文翻译最终修改

英文翻译：基于模糊推理系统的鲁棒性数字图像水印—使用离散余弦变换技术B.Jagadeesha*, P.Rajesh Kumarb, P.Che nna Reddyc摘要：数字水印技术已经成为数字媒体版权保护的最新技术之一。

图像的水印可以在许多方面进行，而对图像水印的一种方法是利用模糊逻辑。

它类似于一个模糊集合的概念，每个元素都可以被一个有序的对，其中一个是值，另一个是隶属函数值。

模糊逻辑系统可以解释不准确的信息，并解释他们的决定。

模糊推理系统是实现模糊逻辑的最简单的方法。

在所提出的方法中，三个模糊推理模型被用来生成嵌入水印和输入到模糊推理系统的权重因子是从人类视觉系统模型。

过程中所使用的性能指标是峰值信噪比，归一化互相关。

该算法是免疫各种图像处理攻击。

关键词：数字水印；离散余弦变换；人类视觉系统；模糊推理系统；归一化互相关性; 峰值信噪比。

1. 引言数字水印尝试隐藏一个消息的真实内容的数字信号。

水印是一个最好的解决方案为数字图像1版权保护。

插入水印，即在一个音频/视频对象的程序被定义为水印。

在水印，水印被添加到在这样一种方式，它仍然在它的封面数据。

这是一个与隐写术密切相关的思想，在一个隐藏的方式中多媒体信号里面的信息。

然而，哪些部分是他们目标2,3。

水印隐藏与数字信号的真实内容相关的消息，而在隐写的多媒体信号与该消息无关，并且它只作为一个庇护，以隐藏其存在，水印主要集中在隐藏的数据的鲁棒性的封面。

嵌入的水印嵌入到宿主图像可以做三个方面，通过将水印信号为原始图像直接第一；我们是嵌入水印的像素直接，这种方法称为空间域方法5。

其次，我们可以将宿主图像值为变换域水印嵌入到组件和组件；这种方法称为变换域的方法，方法第三型为嵌入水印的方法使用，称为混合域6。

每一种方法都有其优缺点，基于应用和嵌入方法选择。

水印可以通过可见的和不可见的方式嵌入水印，这样它就可以被赋予合法所有权的证明。

一些攻击可能会经历，因为数字对象可以处理。

基于 DWT-DCT-SVD 的鲁棒盲视频水印算法

基于 DWT-DCT-SVD 的鲁棒盲视频水印算法陈玉麟;梁栋;张成;鲍文霞【摘要】为更有效地保护多媒体数据，文中提出了一种基于 DWT （discrete wavelet transform）、DCT （discrete cosine transform）与 SVD（singular value decomposition）结合的盲视频水印算法。

利用视频帧内的 R 、G 通道的颜色差值进行关键帧的快速选取，将关键帧的 B 分量进行多级离散小波变换，对变换后的子带进行 Arnold置乱，将水印嵌入到置乱后的子带奇异值中。

当嵌入水印视频受到攻击时，利用彩色图像各颜色通道间像素差值很小和奇异值分解的稳定性，用嵌入水印视频关键帧的 G 分量代替原始视频关键帧的 B 分量，实现水印的盲提取。

实验结果表明，该算法对噪声、滤波、裁剪、帧置乱、帧平均、MPEG （moving pictures experts group）压缩等攻击具有较好的鲁棒性。

%This paper proposed a blind video watermarking algorithm based on DWT ,DCT and SVD .The key frames were selected quickly by color difference of red and green channel in video frames ,then performed the blue component of each of key frames on multi‐level DWT . Sub‐band was scrambled by Arnold transforms ,the watermark was embedded into singular value of scrambled sub‐band .When the video sequence was under attack ,by using the small differences of each color channel pixels of color image and the stability of the singular value decomposition ,this paper used the green component of key frames in a watermarked video to replace the blue component of key frames in the original video and extracted watermark blindly .The experimental results demonstrated that the algorithm wasrobust against noise addition ,filtering attack ,cropping attack ,frame scrambling ,frame averaging and M PEG compression .【期刊名称】《安徽大学学报（自然科学版）》【年(卷),期】2015(000)001【总页数】7页(P25-31)【关键词】视频水印;关键帧;蓝色通道;奇异值;盲提取【作者】陈玉麟;梁栋;张成;鲍文霞【作者单位】安徽大学计算机智能与信号处理教育部重点实验室，安徽合肥230039;安徽大学计算机智能与信号处理教育部重点实验室，安徽合肥 230039;安徽大学计算机智能与信号处理教育部重点实验室，安徽合肥 230039;安徽大学计算机智能与信号处理教育部重点实验室，安徽合肥 230039【正文语种】中文【中图分类】TN911由于互联网技术和数字多媒体技术的快速发展，大量的多媒体数据都可以通过网络被人们轻松的访问、拷贝和传播.因此，寻求有效的方法来解决多媒体数据的版权保护问题日益受到人们重视，其中，数字水印技术成为一种非常有效的版权保护方法［1］.近年来，人们对数字图像水印技术的研究已日趋成熟，而视频水印正处于研究阶段.文献［2］提出了三维Gabor变换的视频水印算法，将水印信息嵌入到三维Gabor变换的系数中，该算法计算复杂度较高，不满足水印的实时性要求.文献［3］基于三维小波变换的空时多分辨率特性，将扩频水印自适应地嵌入到三维小波系数中，鲁棒性较好，但实时性较差且水印为非盲提取.文献［4］中对每4帧图像进行三维小波变换，将水印自适应地嵌入到视频的低频与高频帧中，水印为盲提取且鲁棒性较好.文献［5－6］中的算法对几何攻击都具有较强的鲁棒性，但对于帧置乱、帧丢失等时间同步攻击鲁棒性较弱.文献［7］利用压缩传感和Arnold 变换对水印进行加密，然后将视频关键帧的某一颜色分量进行SVD分解，将加密后的水印嵌入到相应的奇异值中，该算法使水印具有较好的保密性和鲁棒性.文献［8］提出了一种盲视频水印算法，但该算法将视频中的每一帧图像都进行DWT 变换与SVD分解，计算量大且对噪声的鲁棒性较差.文中提出一种基于DWT－DCT－SVD结合的非压缩域视频水印算法，采用灰度图像作为水印.利用帧内R、G通道的颜色差值进行快速关键帧选取.充分利用小波变换的多分辨率特性，对关键帧的B分量进行多级DWT变换，将变换后的高频子带进行Arnold置乱，对置乱后的子带进行DWT和DCT变换，最后将水印嵌入到变换后的子带奇异值中，完成水印的嵌入.同时能对几何步攻击及帧攻击实现水印的盲提取.实验表明，该算法具有较好的不可见性和鲁棒性.1 Arnold变换Arnold变换又称猫脸变换，由于其具有良好的周期性和分散性［9］，能够降低图像像素间的空间相关性，增强水印的安全性，同时能够增强水印的抗剪切能力，因此在数字图像中得到广泛的应用.对于一幅N×N的图像，二维Arnold变换定义如下其中：（x，y）表示原始图像矩阵中像素点的坐标，（x′，y′）表示变换后新图像矩阵中的坐标，且x，y∈｛0，1，2，…，N－1｝.Arnold变换的实质是经过多次迭代后改变原始图像矩阵中所有像素点的空间位置，从而得到一幅置乱后的图像.2 奇异值分解奇异值分解作为一种矩阵分解的方法，在图像压缩、信息隐藏和数字水印等方面得到广泛应用，主要由于其具有以下显著特性［10］：一幅图像的奇异值具有很好的稳定性，当图像受到很小的扰动时，奇异值不会发生太大变化；奇异值对应图像的亮度特性，奇异值向量反映图像的几何特性；利用奇异值矩阵来重构图像时，即使忽略后面很小的奇异值也不影响重构图像的整体质量.从线性代数的角度可以将一幅图像当作一个非负矩阵，记为A∈Rm×n，则大小为m×n的矩阵A的奇异值分解为其中：U∈Rm×m，V∈Rn×n都为酉矩阵.U矩阵的列被称为左奇异值向量，V矩阵的列被称为右奇异值向量S∈Rm×n，为一个对角矩阵，对角线上的元素为奇异值，其主对角线元素满足如下关系由于对角矩阵S中每一个奇异值都为非负值，因此可以将它们作为水印嵌入的区域.3 视频水印算法3.1 水印的预处理为提高水印的安全性，对大小为64×64的原始水印W 进行Arnold变换，得到置乱后的水印W1，置乱效果如图1所示.图1 原始水印及置乱后的水印图像Fig.1 Original watermark and scrambling watermark3.2 关键帧的选取因视频中所含的信息量较大且连续帧之间存在大量的冗余信息，因此文中通过提取视频的关键帧来实现水印的嵌入.根据人眼视觉系统对蓝色最不敏感特性，将水印嵌入到RGB帧的B分量中，因此水印的嵌入对R、G分量不会造成影响.针对传统的基于镜头分割和颜色直方图［11］来选取关键帧，计算复杂度较高，文中提出了一种简单、快速的关键帧选取方法.具体步骤如下：（1）对原始视频进行分帧处理，根据公式（4）计算帧图像R、G颜色分量之间差值的绝对值的和，记为Sk其中：k为帧图像的编号，Rk（x，y，1）、Rk（x，y，2）表示第k帧图像分别在R、G 分量（x，y）处的像素值.（2）从大到小依次选取Sk值对应的前20帧作为关键帧，将选取的关键帧的帧编号位置用密钥key进行保存.3.3 水印的嵌入算法该算法将水印嵌入到关键帧的B分量中，具体的嵌入过程如图2所示.图2 水印嵌入过程Fig.2 Watermark embedding process水印嵌入的主要步骤如下：（1）将视频序列进行分帧处理，选取Sk值较大的前20帧作为水印嵌入的载体. （2）取关键帧的B分量进行2－DWT变换，得到子带LH2.（3）将LH2进行Arnold变换，置乱的迭代次数用key1保存，对置乱后的子带进行DWT变换，得到子带HH3.（4）将HH3进行DWT变换，然后对各子带进行DCT变换，得到的4个子带分别记为LL4、LH4、HL4、HH4，将各子带进行SVD分解，得到各子带的奇异值矩阵Sa1、Sh1、Sv1、Sd1与对应的U、V矩阵.（5）将水印进行Arnold变换，对置乱后的水印W1进行DWT和DCT变换，对变换后的各子带进行SVD分解，得到水印各子带的奇异值矩阵Swa1、Swh1、Swv1、Swd1和对应的Uw、Vw矩阵.（6）按照水印嵌入的加性公式，将水印各子带的奇异值分别嵌入到对应子带LL4、LH4、HL4、HH4的奇异值中其中：Sk 表示Sa1、Sh1、Sv1、Sd1；Swk 表示Swa1、Swh1、Swv1、Swd1；ak 表示不同子带的嵌入强度.（7）对LL4、LH4、HL4、HH4各块进行奇异值重构，再经过DCT逆变换和2级DWT逆变换得到含水印的子带LH2new.（8）将LH2new进行Arnold反置乱，然后进行2－IDWT变换，得到含水印的视频帧图像.（9）按上述步骤将水印重复嵌入到选取的关键帧中，然后将嵌入水印的关键帧放回原始视频中对应位置，得到含水印的视频序列.3.4 水印的提取算法水印提取算法无需原始视频，直接用含水印视频的G分量代替原始视频B分量，将其作为水印提取时的原始数据，为减少G分量受噪声影响，文中用模板为3×3，标准差为0.7的高斯滤波器对其进行滤波.具体的提取过程如图3所示.图3 水印提取过程Fig.3 Watermark extraction process水印提取的具体步骤如下：（1）将待测视频进行分帧处理，按相同算法选取20帧关键帧，从中找到与密钥key中保存的帧编号位置相同的关键帧来进行水印的提取.（2）对待测视频的G分量进行高斯滤波，然后分别将待测视频的B分量和G分量分别进行2－DWT，得到LH2′和LH2＊.（3）分别对LH2′和LH2＊进行与密钥key1相同的Arnold变换，对置乱后的子带进行2级DWT变换.（4）对变换后的各子带进行DCT变换和分解，得到待测视频B分量和G分量的奇异值矩阵分别记为Sn′k和Sg′k.（5）利用下式进行水印奇异值的提取（6）水印各块进行奇异值重构，再对各块进行DCT逆变换和离散小波逆变换，得到完整的置乱水印图像.（7）经Arnold反变换后得到提取后的水印图像.文中用NC（归一化相关系数）值作为水印提取效果的评价标准，定义如下其中：W 表示原始水印；W′表示提取后的水印.4 实验结果与分析选取未压缩的AVI格式的foreman和football视频序列作为测试视频，采用RGB颜色空间，帧率为25f／s，两段视频帧数分别为100帧和173帧，视频帧图像的大小为512×512，水印为64×64的灰度图像.4.1 不可见性用PSNR（峰值信噪比）来评价嵌入水印后视频关键帧的质量.foreman和football视频嵌入水印后所有关键帧的平均PSNR分别为50.13dB和50.67dB，此PSNR值下人眼是无法感知嵌入水印视频帧图像质量下降的，同时图4给出了foreman和football视频嵌入水印前后视频关键帧图像.图4 嵌入水印前后视频关键帧的比较Fig.4 Comparison between the original key frame and watermarked key frame4.2 鲁棒性分析一般对视频水印的攻击可分为两类：一类是一般攻击方式，如噪声、滤波、旋转、缩放、JPEG压缩等攻击；另一类是针对视频的帧攻击方式，如帧置乱、帧丢失、帧平均、MPEG压缩等攻击.为验证文中视频水印算法的有效性，对水印的鲁棒性进行如下测试，并给出相应的实验结果.4.2.1 一般攻击的鲁棒性分析从嵌入水印视频和原始视频的帧编号位置相同的关键帧中取1帧作为测试样本，表1给出了该算法在一般攻击下从foreman和football视频中提取的水印NC值，图5给出了football视频在不同攻击下重构水印图像.表1 一般攻击下该算法提取的水印NC值Tab.1 The algorithm extracted watermark NC values under the general attacks攻击方式及参数 NC值foreman视频 NC值football 0.964 8 0.974 3椒盐噪声（var＝0.2） 0.975 2 0.959 7高斯滤波（9×9σ＝0.5） 0.999 0 0.998 7中值滤波（3×3） 0.999 0 0.999 2均值滤波（3×3） 0.998 9 0.998 8锐化（80） 0.997 5 0.998 8旋转（10°） 0.831 7 0.824 9裁剪1／3 0.996 3 0.996 3裁剪1／2 0.984 40.988 8放大2倍再还原 0.999 8 0.999 9缩小2倍再还原 0.999 5 0.999 4 JPEG压缩 Q＝70% 0.925 2 0.942 4 JPEG压缩视频高斯噪声（var＝0.1）Q＝40% 0.907 1 0.923 8图5 football视频在不同攻击下重构的水印图像Fig.5 Football video reconstructed watermark image under different attacks为验证该算法的有效性，保证试验在尽可能相同的情况下进行，文中选取64×64大小的二值图像作为水印，为提高水印的提取效果，需先对水印进行二极化处理.取foreman视频作为水印的载体数据，表2给出了该算法与文献［12］的比较结果，同时图6给出了该算法在不同攻击下水印重构的结果.表2 该算法与文献［12］的比较结果Tab.2 Comparison between NC values obtained by this algorithm and reference［12］攻击方式及参数该文算法文献［12］0.998 2 0.913 5椒盐噪声（var＝0.1） 0.990 8 0.862 7高斯滤波（9×9σ＝0.5） 0.999 1 0.968 8中值滤波（3×3） 0.999 6 0.970 4亮度增加1.5倍 0.987 4 0.928 1旋转（1°） 0.879 6 0.905 9裁剪1／4 0.997 9 0.882 7裁剪1／2 0.985 4 0.732 8放大2倍再还原 0.999 9 0.987 7缩小2倍再还原0.999 5 0.900 1 JPEG压缩算法高斯噪声（var＝0.01）Q＝40% 0.923 7 0.857 4图6 该算法在不同攻击下重构的水印图像Fig.6 This algorithm reconstructed watermark image under different attacks由表2可知，论文算法对噪声、滤波、剪切、亮度变化、缩放和JPEG压缩等攻击明显优于文献［12］.在抗旋转攻击能力方面较弱，但在较小角度旋转的情况下提取的水印还是能够清晰地辨别.4.2.2 帧攻击的鲁棒性分析视频的帧攻击方式是目前视频应用的主要问题，由于帧置乱、帧丢失、帧平均等攻击能破坏视序列的时间同步性，从而使嵌入的水印无法被检测或提取.从选取的关键帧中取1帧作为测试样本，图7的（a）、（b）、（c）给出了该算法在帧攻击下从foreman和football视频中提取的水印 NC值，图（d）、（e）、（f）分别为在不同的帧攻击下从foreman视频中重构的水印图像.由图7可见，该算法对于帧置乱、帧丢失、帧平均具有较强的鲁棒性.由于水印嵌在B分量中，对帧的R、G分量值无干扰，因此利用文中关键帧提取算法能够将含水印的帧检测出来，并从Sk值较大的关键帧中提取水印.MPEG压缩攻击是视频水印较常见的攻击方式，该文将foreman和football的含水印视频以1.2 Mbps的码率进行压缩，提取后水印NC值分别为0.970 1和0.960 0.该结果表明该算法对于MPEG压缩也具有较强的鲁棒性.图7 foreman视频在帧攻击下重构的水印NC值Fig.7 Foreman video reconstructed watermark image under frame attacks5 结束语为满足水印的实时性与盲提取，文中提出了基于DWT、DCT与SVD结合的鲁棒盲视频水印算法.只利用视频帧内R、G通道颜色差值选取关键帧，将关键帧进行2－DWT变换，然后对子带进行Arnold变换，利用奇异值分解的性质和小波变换的多分辨率特性，将水印嵌入到置乱后的子带奇异值中，且当视频受到不同攻击时能实现水印的盲提取.实验表明，该算法对视频水印的多种攻击具有较强的鲁棒性，但水印提取过程中需要用到部分水印数据.因此，下一步主要研究工作是在不使用原始水印数据的情况下进行水印的盲提取.参考文献：［1］Podilchuk C I，Delp E J.Digital watermarking：algorithms and applications［J］.IEEE Signal Processing Magazine，2001，18（4）：33－46.［2］张立和，伍宏涛，胡昌利.基于三维 Gabor变换的视频水印算法［J］.软件学报，2004（8）：1252－1258.［3］李英，高新波，姬红兵.一种基于三维小波的视频水印空时算法［J］.系统工程与电子技术，2005（1）：16－19.［4］霍菲菲，高新波.基于三维小波变换的视频水印嵌入与盲提取算法［J］.电子与信息学报，2007（2）：447－450.［5］杨晓元，钮可，魏萍，等.一种抗几何攻击的视频水印算法［J］.计算机工程，2007（8）：142－144.［6］楼偶俊，王相海，王钲旋.抗几何攻击的量化鲁棒视频水印技术研究［J］.计算机研究与发展，2007（7）：1211－1218.［7］Jyothish L G，Veena V K，Soman K P.A cryptographic approach to video watermarking based on compressive sensing，arnold transform，sum of absolute deviation and SVD［C］／／Emerging Research Areas andInternational Conference on Microelectronics，Communications and Renewable Energy（AICERA／ICMiCR），2013：1－5.［8］Rajab L，Al－Khatib T，Al－Haj A.Hybrid DWT－SVD video watermarking［C］／／International Conference on Innovation in Information Technology，2008：588－592.［9］Wu L，Zhang J，Deng W，et al.Arnold transformation algorithm and anti－arnold transformation algorithm［C］／／International Conference on Information Science and Engineering（ICISE），2009：1164－1167. ［10］Liu R，Tan T.An SVD－based watermarking scheme for protecting rightful ownership［J］.IEEE Transactions on Multimedia，2002，4（1）：121－128.［11］刘政凯，汤晓鸥.视频检索中镜头分割方法综述［J］.计算机工程与应用，2002，23：84－87.［12］毛运柳，黄东军.基于SURF视频分割的视频水印算法［J］.计算机工程，2010，19：241－243.。

基于DWT-SVD和Turbo码的彩色图像盲水印算法

基于DWT-SVD和Turbo码的彩色图像盲水印算法朱建忠;姚志强【期刊名称】《吉林大学学报（理学版）》【年(卷),期】2014(000)004【摘要】为进一步提高数字水印系统的性能，提出一种离散小波变换域中基于奇异值分解和Turbo 纠错码的彩色图像盲水印算法。

首先对数字水印进行混沌加密预处理，然后进行Turbo 编码，再把编码后的水印信息嵌入到彩色图像蓝色分量小波分解后低频子带的奇异值矩阵中。

提取水印时不需原始载体图像的参与，更便于实际应用。

仿真实验结果表明，该水印算法不可见性较好，且对常见攻击和几何攻击具有较好的稳健性。

%In order to improve the performance of digital watermarking system,a watermarking algorithm was proposed for color images based on the combination of discrete wavelet transform, singular value decomposition and Turbo codes.Firstly,digital watermarking was pretreated by chaotic encrypting and Turbo coding;secondly,the blue component of color image was decomposed via discrete wavelet transform;finally,the pretreated watermarking information was embedded into the low frequency band of the blue component which had been DWT of the original image by modifying its singular value.It does not need the original image when the watermarking is extracted, so it can be applied in the practice easily.Simulation results show that the watermarking algorithm has good invisibility and robustness to common attacks and geometric attacks.【总页数】6页(P773-778)【作者】朱建忠;姚志强【作者单位】福建广播电视大学电子信息与计算机系，福州 350003;福建师范大学软件学院，福州 350108【正文语种】中文【中图分类】TP391【相关文献】1.基于DWT-SVD和Fibonacci变换的彩色图像盲水印算法 [J], 蔡宜嘉;牛玉刚;苏庆堂2.一种基于Turbo码与SVD的彩色图像双水印算法 [J], 李文;李宏昌;3.基于非方阵图像DWT-SVD的盲水印算法研究与仿真 [J], 张帅; 杨雪霞4.基于QR码与混沌加密的Contourlet域彩色图像盲水印算法 [J], 季诺然;吕晓琪;谷宇;赵瑛;刘坤5.基于QR码和DWT-SVD技术的双彩色盲水印算法 [J], 孙业强;王晓红;李肖赫因版权原因，仅展示原文概要，查看原文内容请购买。

基于DWT低频系数的自适应盲水印算法

基于DWT低频系数的自适应盲水印算法
应隽;刘勍
【期刊名称】《自动化与仪器仪表》
【年(卷),期】2011(0)3
【摘要】提出了一种基于小波变换低频系数的自适应数字水印算法。

通过小波变
换将图像分解成不同空间与频率的系数,这些系数之间存在着一种树状结构关系。

利用这种树状结构关系将低频系数分成两类,一类对应图像的强纹理区域,一类对应
图像的弱纹理区域;针对分类的结果将不同强度的水印分量嵌入到对应的低频小波
系数中。

该算法具有很好的不可感知性,实验结果表明该算法有较好抗JPEG压缩、低通滤波、剪切等各种攻击的能力。

【总页数】3页(P132-134)
【关键词】小波变换;数字水印;Arnold变换
【作者】应隽;刘勍
【作者单位】合肥工业大学计算机与信息学院;天水师范学院物理与信息科学学院【正文语种】中文
【中图分类】TP13
【相关文献】
1.基于DWT域自适应均值量化的彩色图像盲水印算法 [J], 易云
2.基于小波域低频系数的自适应盲视频水印算法 [J], 任克强;张凯;谢斌
3.一种基于DWT和DCT的自适应盲水印算法 [J], 于瑞琴
4.一种基于DWT的自适应扩频盲检数字水印算法 [J], 陆萍
5.基于ICA和DWT的自适应盲水印算法 [J], 赵伟;陈仁安;张晓玲;游荣义
因版权原因，仅展示原文概要，查看原文内容请购买。

基于DWT的彩色图像鲁棒盲水印新算法

基于DWT的彩色图像鲁棒盲水印新算法
莫红飞;孙德辉
【期刊名称】《电脑知识与技术》
【年(卷),期】2007(003)014
【摘要】本文提出了一种基于小波域低频系数量化的彩色图像鲁棒盲水印新算法.首先将RGB彩色宿主图像转换为YCbCr颜色空间,然后对亮度分量Y进行三层小波分解,采用适当的量化间隔对低频逼近子带LL3进行均值量化,根据量化系数的奇偶特性再与二值图像置乱转化后的水印序列进行异或运算,从而把水印嵌入到亮度分量的逼近子带中.仿真实验结果表明,本文提出的算法对JPEG/JPEG2000压缩、添加噪声、滤波、剪切、像素移位等多种攻击有较强的鲁棒性,透明性好,并且能实现水印的盲提取.
【总页数】3页(P540-541,580)
【作者】莫红飞;孙德辉
【作者单位】华东电子工程研究所,安徽,合肥,230031;合肥工业大学,计算机与信息学院,安徽,合肥,230009
【正文语种】中文
【中图分类】TP18
【相关文献】
1.结合DWT和SVD的鲁棒盲水印算法 [J], 胡娟;杨格兰;严权锋
2.一种鲁棒的基于DWT域自适应量化步长的图像盲水印算法 [J], 张专成;张殿富;
闫小萍
3.基于DWT的彩色图像鲁棒盲水印新算法 [J], 莫红飞;孙德辉
4.基于DWT-SVD和Turbo码的彩色图像盲水印算法 [J], 朱建忠;姚志强
5.基于DWT域的多通道彩色图像盲水印算法 [J], 刘志军
因版权原因，仅展示原文概要，查看原文内容请购买。

基于DWT-SVD和Fibonacci变换的彩色图像盲水印算法

基于DWT-SVD和Fibonacci变换的彩色图像盲水印算法蔡宜嘉;牛玉刚;苏庆堂【期刊名称】《计算机应用研究》【年(卷),期】2012(029)008【摘要】为提高水印鲁棒性,将离散小波变换(DWT)、奇异值分解(SVD)和斐波纳契(Fibonacci)变换结合,提出一种新的算法.首先,用Fibonacci变换对拟嵌入的水印进行置乱处理；然后,对宿主彩色图像R、G、B三个分量进行二级小波变换和基于4x4分块的奇异值分解,并用混沌序列选择若干对子块；最后,根据人类视觉系统(HVS)特性对三个分量分配嵌入量、确定嵌入强度,并通过修改每对子块最大奇异值来实现水印嵌入.实验结果表明本方案具有良好的水印不可见性和鲁棒性.%To improve the robustness of digital watermark, this paper proposed a novel robust watermarking algorithm for color images based on discrete wavelet transformation ( DWT) , SVD and Fibonacci transformation. Firstly, this algorithm adopted the Fibonacci transformation to scramble the original watermark before embedded. Then,it decomposed the R,G,B components of the host color image by 2-level-DWT, and partitioned the obtained sub-bands into non-overlapping blocks of 4×4 pixels and then applied SVD to the blocks. Meanwhile it used the chaotic sequence to select several couples of blocks. Finally, it determined the embedding capacity and the intensity of each component according to the characteristics of human vision system to meet the requirements of invisibility and embedded one bit of watermark by modifying the relative size of the first singular valuesof one couple blocks. Experimental results show that the proposed algorithm has a good performance in imperceptibility and robustness of watermark.【总页数】4页(P3025-3028)【作者】蔡宜嘉;牛玉刚;苏庆堂【作者单位】华东理工大学化工过程先进控制和优化技术教育部重点实验室,上海200237;华东理工大学化工过程先进控制和优化技术教育部重点实验室,上海200237;华东理工大学化工过程先进控制和优化技术教育部重点实验室,上海200237【正文语种】中文【中图分类】TP393.04【相关文献】1.基于Arnold变换和DWT彩色图像数字盲水印算法 [J], 王启亮;柏逢明2.基于 DWT-SVD 和 Arnold 变换图像盲水印算法 [J], 李林静;郑月斋3.基于整型小波变换的彩色图像盲水印算法 [J], 苏庆堂4.基于DWT-SVD和Turbo码的彩色图像盲水印算法 [J], 朱建忠;姚志强5.基于DCT-DQFT变换和QR分解的彩色图像盲水印算法 [J], 马玲;覃亮成因版权原因，仅展示原文概要，查看原文内容请购买。