基于神经网络的鲁棒型图像水印算法研究

发布时间：2018-03-29 11:50

本文选题：数字水印　切入点：人工神经网络　出处：《中国地质大学(北京)》2017年硕士论文

【摘要】：信息技术的迅猛发展极大得丰富了人们的生活,越来越多的数字作品以多媒体的形式呈现在了互联网上,在方便人们交流与学习的同时,却也带来了许多的信息安全问题,一些盗版问题和版权纠纷日益严重。数字水印是一种有效的数字产品保护和数据安全维护技术,它将版权信息以特定的形式隐藏到相关的数字作品中,用以证明创作者对其作品的所有权,从而成为一种知识产权保护和数字多媒体防伪的有效手段。本文主要研究人工神经网络在数字图像水印技术中的应用。人工神经网络是一种在生物神经网络的启发下建立的数据处理模型,主要通过调整神经元之间的权值来对输入的数据进行建模,具有高度的非线形全局拟合能力和容错能力,以及很好的自适应自学习功能。本文在此基础上提出了两种数字图像盲水印算法。一种是图像空域上的水印算法,该算法根据灰度共生矩阵的特征量选取图像的局部纹理复杂区域嵌入水印信息,提高了水印的不可见性;同时利用人工神经网络构建原始图像上像素间的关系模型,实现了水印信息的盲提取。另外一种是小波变换域上的水印算法,该方法利用多尺度分析研究图像的变换域特性,将水印信息嵌入到图像低频区域的奇异值系数中,以达到增强水印鲁棒性的效果。从实验的结果和数据来看,以上两种方法都具有良好的不可见性。本文的空域算法对中高品质的JPEG2000压缩、噪声和高斯滤波攻击均具有不错的抗性,但对于中值滤波和均值滤波则会失效;相比之下,本文的小波变换域算法从整体上来说抗干扰能力更强,对于文中测试的各种JPEG2000压缩、噪声、滤波、裁剪攻击都具有很强的鲁棒性。
[Abstract]:The rapid development of information technology has greatly enriched people's lives. More and more digital works are presented on the Internet in the form of multimedia. While it is convenient for people to communicate and learn, it also brings a lot of information security problems. Some piracy problems and copyright disputes are becoming more and more serious. Digital watermarking is an effective technology for digital product protection and data security maintenance. It hides copyright information in specific forms in related digital works. To prove the author's ownership of his work, In this paper, we mainly study the application of artificial neural network in digital image watermarking technology. Artificial neural network is a kind of biological neural network. The established data processing model, The input data is modeled by adjusting the weights between neurons, which has a high ability of global fitting and fault tolerance. Based on this, two digital image blind watermarking algorithms are proposed, one is image watermarking algorithm in spatial domain, the other is image spatial watermarking algorithm. According to the characteristic quantity of gray level co-occurrence matrix, the algorithm selects the local texture complex region of the image to embed the watermark information, improves the invisibility of the watermark, and constructs the relationship model of pixels on the original image by using artificial neural network. The other is the watermarking algorithm based on wavelet transform, which uses multi-scale analysis to study the characteristics of image transform domain, and embeds the watermark information into the singular value coefficients of the low frequency region of the image. In order to enhance the robustness of watermark, the experimental results and data show that the above two methods are invisible. Both noise and Gao Si filtering attack have good resistance, but they will fail for median filter and mean filter. In contrast, the wavelet transform domain algorithm in this paper has stronger anti-jamming ability on the whole, for various JPEG2000 compression tested in this paper, Noise, filtering and clipping attacks all have strong robustness.
【学位授予单位】：中国地质大学(北京)
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP309.7

【相似文献】