DCT域图像水印的局部最优非线性检测

发布时间：2018-04-05 05:22

本文选题：α稳定分布　切入点：局部最优检测　出处：《曲阜师范大学》2014年硕士论文

【摘要】：网络的开放性和资源共享给网络信息安全带来了极大的隐患（如侵权、篡改等等）。因此，多媒体作品的版权保护问题亟待解决。近年来，一种用于知识产权保护的新方法越来越流行，即在多媒体信息中嵌入数字水印。数字水印技术是通过某种算法在多媒体数据中嵌入特定的信息，具有三大显著特征：不可见性、鲁棒性和安全性。目前绝大多数检测算法采用了线性相关的方法，由信号检测的基本理论可知，基于线性相关的水印检测方法只有在水印载体服从高斯分布时才是最优的。研究结果表明，在数字图像的时/空域或者变换域，以高斯分布来对载体图像进行统计建模是不合适的。因此，，从水印检测的角度来看，线性相关水印检测方法没有考虑到载体图像的实际统计分布特性，其优化条件不复存在，检测性能也随之严重退化。对于离散余弦变换域（DCT）图像水印，数据呈重尾分布，相关检测显然不是最优的检测方案。不可感知性是数字水印的一个基本特征，这就决定了水印信号的检测是一个弱信号的检测问题。研究表明非线性接收器特别适合呈重尾分布的噪音弱信号的检测，是局部最优检测方案。这促使了局部最优柯西非线性检测和零记忆非线性检测在DCT变换域图像中的应用。本文主要工作如下：首先，介绍DCT图像水印相关理论基础知识，包括DCT域图像水印的生成、嵌入和检测结构，为下面研究DCT域图像水印的检测算法做铺垫。其次，对几种常见的DCT系数模型进行了分析，并提出一种新的数据建模方法，即均衡稳定簇模型，并呈现使用稳定分布的方法对图像的DCT系数进行建模的结果。然后，提出了两种局部最优非线性检测算法，即局部最优柯西非线性检测和零记忆非线性检测。通过计算似然比，对检测性能进行理论分析。最后，对提出的两种非线性检测算法进行量化攻击，通过计算似然比，对检测性能进行理论分析，并通过仿真实验绘制接收机工作特性曲线(ROC)，对检测器的性能及鲁棒性进行比较分析。
[Abstract]:The openness and resource sharing of network bring great hidden trouble (such as infringement, tampering, etc.) to network information security.Therefore, the copyright protection of multimedia works needs to be solved.In recent years, a new method for intellectual property protection is becoming more and more popular, that is, embedding digital watermarking into multimedia information.Digital watermarking technology is to embed specific information in multimedia data through some algorithm, which has three characteristics: invisibility, robustness and security.At present, most detection algorithms adopt linear correlation method. From the basic theory of signal detection, the linear correlation based watermarking detection method is optimal only when the watermark carrier is distributed from Gao Si.The results show that it is not appropriate to use Gao Si distribution to model the carrier image in the time / spatial domain or transform domain of the digital image.Therefore, from the point of view of watermark detection, linear correlation watermarking detection method does not take into account the actual statistical distribution of the carrier image, its optimization conditions no longer exist, and the detection performance is seriously degraded.For DCT image watermarking in discrete cosine transform domain, the data is heavy-tailed, and correlation detection is obviously not the optimal detection scheme.Imperceptibility is a basic feature of digital watermarking, which determines that the detection of watermark signal is a weak signal detection problem.It is shown that the nonlinear receiver is especially suitable for the detection of noise weak signals with heavy-tailed distribution and is a local optimal detection scheme.This results in the application of local optimal Cauchy nonlinear detection and zero memory nonlinear detection in DCT transform domain images.The main work of this paper is as follows:First of all, the basic theory of DCT image watermarking is introduced, including the generation, embedding and detection structure of DCT domain image watermarking, which pave the way for the following research of DCT domain image watermarking detection algorithm.Secondly, several common DCT coefficient models are analyzed, and a new data modeling method, equilibrium stable cluster model, is proposed, and the results of modeling the DCT coefficients of images by using the stable distribution method are presented.Then, two local optimal nonlinear detection algorithms are proposed, that is, local optimal Cauchy nonlinear detection and zero-memory nonlinear detection.The detection performance is theoretically analyzed by calculating likelihood ratio.Finally, the proposed two nonlinear detection algorithms are quantitatively attacked, and the detection performance is theoretically analyzed by calculating likelihood ratio.The performance and robustness of the detector are compared and analyzed by drawing the operating characteristic curve of the receiver through simulation experiments.
【学位授予单位】：曲阜师范大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP309.7

【参考文献】