《基于小波分析和神经网络的图像数字水印技术研究》由会员分享，可在线阅读，更多相关《基于小波分析和神经网络的图像数字水印技术研究（94页珍藏版）》请在金锄头文库上搜索。
1、西安科技大学 硕士学位论文 基于小波分析和神经网络的图像数字水印技术研究 姓名：张蒲临 申请学位级别：硕士 专业：应用数学 指导教师：乔宝明 2011 论文题目：基于小波分析和神经网络的图像数字水印技术研究 专 业：应用数学 硕 士 生：张蒲临 (签名) 指导教师：乔宝明 (签名) 摘 要 数字水印技术是目前信息安全技术领域的一个新方向， 是一种可以在开放的网络环 境下保护数字作品版权和认证来源及完整性的新型技术。 根据水印的隐藏位置可以把水 印算法分为空域算法和变换域算法两大类。空域水印算法与变换域水印算法相比，优点 是计算速度比较快，时间复杂度比较低，具有较大的信息隐藏量等特性；缺点是此种
2、算 法实现的数字水印比较脆弱，鲁棒性较差。变换域水印算法与空域水印算法相比，优点 是在变换域中嵌入的水印信号能量可以分布到空域的所有像素上， 有利于提高数字水印 的不可见性和鲁棒性；缺点是变换域水印算法一般计算量比较大。 小波变换在数字水印技术中的应用主要体现在图像数字水印方面， 其基本思想是把 图像进行多分辨率分解，生成不同空间和独立频带的子图像，然后利用小波系数的局部 特性实现感兴趣区域的水印嵌入。神经网络在数字水印技术中的应用可分为两个方面： 一是在水印嵌入时使用神经网络对图像进行分类或产生自适应于图像的水印， 其目的是 提高水印的嵌入强度和图像的保真度；二是使用神经网络进行水印检测，其
3、目的是提高 水印检测的正确率。本文根据小波分析和神经网络的特点，提出了一种基于小波包变换 及Hopfield网络的变换域水印算法。 本文主要工作有： (1) 对小波分析、人工神经网络和数字水印的基本理论做了简单阐述，深入研究了 三种数字水印算法。 (2) 提出了一种基于小波包变换及 Hopfield 网络的水印算法。该水印算法利用小波 包变换， 选择中低频嵌入水印信息， 实现了特定频带的水印嵌入， 以此来抵抗滤波攻击； 利用离散余弦变换使子图像能量更加集中，以此来有效抵抗旋转攻击；利用 Hopfield 神 经网络进行了水印的检测， 不仅提高了水印检测的正确率， 而且增强了水印的提取效果。 实
4、验结果表明该算法能较好的抵抗裁剪攻击、压缩攻击；能抵抗一定量的噪声攻击和缩 放攻击；对于滤波攻击和旋转攻击具有较强的鲁棒性。 (3) 本文对提出的水印算法和其它三种水印算法进行了六种水印攻击实验。实验结 果表明本文提出的水印算法在抗压缩攻击、噪声攻击、滤波攻击、旋转攻击中，效果优 于其它三种水印算法。 (4) 本文尝试对提出的水印算法用粒子群算法优化水印嵌入强度参数。在实验中发 现该方法只能对具体问题的具体环节中的参数进行优化，并且优化的参数不具有通用 性。今后可在嵌入强度参数优化方面做进一步的研究。 关 键 词：数字水印；离散小波变换；神经网络；离散余弦变换；粒子群算法 研究类型：应用研究
5、Subject : The Technology Study Of Picture Digital Watermarking Based On Wavelet Analysis And Neural Network Specialty : Applied Mathematics Name : Zhang Pulin (Signature) Instructor : Qiao Baoming (Signature) ABSTRACT Digital watermarking technology is a new direction of the information security fie
6、ld,is a new technology to protect the digital copyright,to authenticate digital information sources and integrity in an open network environment.Watermarking algorithm is divided into spatial and transform domain algorithms according to the hidden location of the watermark.In contrast with transform
7、 domain watermarking algorithms,the advantage of spatial domain watermarking algorithms is computationally faster, and time complexity is relatively low,with a large amount of hiding information etc.Drawback of spatial domain watermarking algorithms is fragile digital watermarking,robustness is poor
8、.In contrast with spatial domain watermarking algorithms,the advantage of transform domain watermarking algorithms is that embedded watermark signal energy in the transform domain can be distributed to all the pixels on the airspace,which will help improve digital watermark invisibility and robustne
9、ss.The disadvantage of the general transform domain watermarking algorithms is larger in computation. Wavelet transform in digital watermarking technology is mainly reflected in the application of digital image watermarking.The basic idea is to decompose the multi-resolution image to generate differ
10、ent spatial and independent sub-band images,to use the local characteristics of wavelet coefficients to achieve watermark embedding in region of interest.Neural Networks in the application of digital watermarking technology can be divided into two aspects.On the one hand,the use of neural network in
11、 watermark embedding to classify images or generate adaptive watermark on images,which aims to improve the embedding strength and the image of the security truth degree.On the other hand, using neural network in watermark detection,which aims to improve the accuracy of watermark detection.In the con
12、text of wavelet analysis and artificial neural networks,this paper presents a watermarking algorithm based on wavelet packet transform and Hopfield networks. The main research of this paper as followings: (1) This paper briefly describes the basic theory of the wavelet analysis、artificial neural net
13、work and digital watermarking,and studies three kinds of digital watermarking algorithms. (2) This paper presents a watermarking algorithm based on wavelet packet transform and Hopfield networks.The watermarking algorithm uses wavelet packet transform to select the low-frequency to embed watermark,
14、to achieve a specific frequency band to embed watermark,in order to resist the filtering attack;The watermarking algorithm uses discrete cosine transform to make the sub-image energy more concentrated, in order to be effective against spin attack;The watermarking algorithm uses Hopfield neural netwo
15、rk to detect the watermark,not only improves the accuracy of watermark detection,but also enhances the watermark extraction.The results show that the algorithm can resist cutting attack and compression attack;can resist a certain amount of noise attacks and scaling attacks;for filtering attack and s
16、pin attack the algorithm has strong robustness. (3) This paper carries out six watermark attack experiments on the proposed watermarking algorithm and the other three watermarking algorithms.Experimental results show that the effect of the proposed watermarking algorithm is better than the other three watermarking algorithms against compression
