【作者】 胡玲娜；

【导师】 蒋铃鸽；

【作者基本信息】 上海交通大学 ， 通信与信息系统， 2010， 博士

【摘要】 静止图像数字水印的盲检测技术在对原始载体信息、水印信息、水印嵌入提取方法与密码等基本条件一无所知的情况下,对疑似图像实施水印存在性及其相关信息的盲检测。本论文针对静止图像不同类型数字水印算法提出多个盲检测算法,并对所提出算法进行详细的性能分析和计算机仿真验证。主要研究内容及成果如下:第一,研究针对静止图像最低有效位(LSB)数字水印的盲检测算法。提出两种静止图像的盲检测算法。基于统计特性的盲检测算法借助图像工程理论,针对LSB替换嵌入算法,利用像素及其周围像素间的拉普拉斯值,推导了待测图像中水印信息的长度计算公式。算法对集中嵌入与分散嵌入的水印信息均有较好的检测效果,低嵌入率时仍能保证较高检出率。基于图像分割的盲检测算法利用水印实际嵌入的局部性,对待测图像进行分割,研究区域最小生成树中节点的度的分布,根据Fisher鉴别准则对待测图像实现数字水印盲检测。算法对替换嵌入与匹配嵌入的水印信息检测效果良好。第二,研究针对静止图像离散余弦变换(DCT)数字水印的盲检测算法。提出基于随机共振理论的盲检测算法。算法利用随机共振原理,通过吸收外界能量补充待测信息能量,放大DCT系数统计参数在水印嵌入前后的变化,选择其中变化程度较大的参数构成样本矢量,为使用支持向量机(SVM)算法实现盲检测做准备。最后在SVM算法的基础上,对集中嵌入与分散嵌入的两种常见DCT数字水印实现盲检测。仿真结果表明,该算法在嵌入强度较低时能够保证较好的检测效果。第三,研究针对静止图像离散小波变换(DWT)数字水印的盲检测算法。提出基于核Fisher鉴别分析(KFDA)理论的盲检测算法。算法借鉴静止图像DCT数字水印盲检测的研究成果,首先将待测信息通过随机共振系统,对其中微弱信号进行放大。然后计算并选取能够表征待测信息统计特征的样本矢量,最后选择KFDA理论实现盲检测。理论分析和仿真结果表明,算法有效提高了低嵌入强度下数字水印的盲检测率,同时大幅降低了虚警率。第四,研究基于数字水印及其检测理论的手持数字视频广播(DVB-H)网络发射机识别算法。其中空域识别算法将数字水印信号直接嵌入DVB-H发射机中,接收端通过检测嵌入的水印信息识别发射机。算法简单易实现,误码率(BER)性能显示由水印信息引入的失真可以忽略不计。频域识别算法将水印信息嵌入在DVB-H系统频域信号中,接收端以迭代反馈方式提取水印信号,识别发射机状态。频域识别算法较大程度改善了空域识别算法对系统的依赖性。根据水印检测理论,两种算法在嵌入强度较低的无线环境中均能检测出嵌入信息,并一定程度提高发射器的峰均比(PAPR)性能。尽管添加了额外信息,频谱利用率保持不变。更多还原

【Abstract】 The technology of blind detection on the digital watermarking of still image needs to work without any necessary information, such as the original carrier information, watermarks information, schemes and key of embedding or extraction etc. The technology is usually used to detect existence of watermarks or some watermarks information. The dissertation proposed some blind detection schemes aiming at different type watermarking. Detailed analysis on properties and computer simulations are given for every scheme. The main topics of this dissertation are as the follows:Firstly, two algorihms of blind detection on LSB (Least Significant Bit) digital watermarking of still image are proposed. With the aid of image engineering, one algorithm based on statistic property uses the Laplace value of the pixel and other pixels around it to deduce length formula of watermarks in the test image. The algorithm has good detection results on both concentrated and dispersed watermarking information. It also could keep high detection probability at low embedding rate. The algorithm based on image segmentation utilizes the locality property of actual watermarking to segment the test images. It studies the distribution of node degree in MSTs. It carries out blind detection on digital watermarking according to Fisher principle. The algorithm has good detection results on the information both of replacement embedding and matching embedding.Secondly, the algorithm of blind detection on DCT (Discrete Cosine Transformation) digital watermarking of still image is proposed. By energy of noise, the stochastic resonance system is utilized to increase amplitude of watermarking information greatly. As a result, the difference between stego watermarking information and original information increases. Based on SVM algorithm, the blind detection is carried out on two general DCT watermark embedding methods which are concentrative embedding algorithm and dispersed embedding algorithm. Simulation results show high detection probability at low embedding strength.Thirdly, the algorithm of blind detection on DWT (Discrete Wavelet Transform) digital watermarking of still image is proposed. The algorithm utilizes KFDA (Kernel Fisher Discrimination Analysis) theory. With the help of research results of blind detection on DCT digital watermarking, the algorithm passes the test information by stochastic resonance system so as to amplify weak signals. Then the algorithm chooses suitable sample vector by computation. KFDA theory which is a kind of learning machine with high precision is used to realize blind detection. Both theoretical analysis and simulation results show that the algorithm improves detection probability at low embedding strength. At the same time the algorithm also decreases false alarm rate.Finally, algorithms of transmitter identification (TxID) method for DVB-H(Digital Video Broadcasting-Handheld) system are proposed. They are both based on digital watermarking technology. The spatial embedding algorithm embeds digital watermarking signals in the time domain of DVB-H transmitter. Receiver identifies the transmitter by detecting the embedded information. The algorithm is easy to be realized. The performance of BER shows that the distortion induced by digital watermarking information can be ignored. While the frequency embedding algorithm signals are embedded in the frequency domain of transmitter. Receiver identifies the transmitter by detecting the embedded information with the method of iteration and feedback. Compared with the spatial embedding algorithm, the frequency one improves dependence on system. According the theory of watermarking, two algorithms proposed can detect watermarking information even in wireless environment with low embedding rate. And the algorithms can improve PRPA of transmitter in some way. Although extral information is embedded, utilization rate of frequency spectrum is unchanged.更多还原