自适应图像数字水印研究-智能与分布计算实验室

自适应图像数字水印研究

姓名	毛万欣
论文答辩日期	2005.05.11
论文提交日期	2005.05.12
论文级别	硕士
中文题名	自适应图像数字水印研究
英文题名	Research on Adaptive Image Digital Watermarking
导师1	卢正鼎
导师2
中文关键词	数字水印;鲁棒性;不可感知性;自适应
英文关键词	Digital Watermarking;Robustness;Imperceptibility;Adaptive
中文文摘	随着数字多媒体技术以及计算机网络的飞速发展，数字作品的复制、修改和传输变得十分容易，数字作品的版权保护问题变得日益突出。数字水印技术是数字多媒体版权保护的新技术，对于该技术的研究还处于起步阶段，关于数字水印的基本原理、关键技术以及应用开发等问题还有待于系统深入的研究。正确合理的系统框架是对数字水印技术进行研究的保证。剖析了数字水印的基本框架，研究了框架整体的特性和评价标准，并对各个组成部分的功能特性、技术特点进行了分析。对数字水印算法所追求的鲁棒性、不可感知性和安全性，进行了理论的分析，深入研究了提高这些特性的方法。数字水印的鲁棒性和不可感知性天生就是矛盾的，想要实现两者兼顾是一个难题。对以利用图像内容和视觉特性的自适应算法进行了深入的研究分析，论述了其产生的必然性、实质和优点。对基于离散余弦变换的自适应算法进行了改进。分别计算低频区系数和中频区系数的平均值，选择介于这两个平均值之间的系数作为水印的嵌入位置；在水印嵌入后利用计算感知距离的视觉模型计算出载体图像和嵌入水印后图像的感知距离，将此感知距离与预定义的阈值进行比较，如果小于阈值嵌入完成，如果大于阈值则要调整水印的嵌入强度，并以新的嵌入强度重新进行水印的嵌入。这样既保证水印有足够的鲁棒性也充分保证了水印的不可感知性。对基于离散小波变换的自适应算法进行了改进。利用小波域视觉模型充分考虑图像内容和人类视觉特性计算出每个小波系数处的临界可见差异，用此临界可见差异乘以对应子带的拉伸因子作为水印的嵌入强度，使水印的嵌入强度更精确；用水印信息与待嵌入水印的小波系数的比值与预先定义的阈值进行比较，如果小于阈值则选取该小波系数进行水印的嵌入，采用这种用水印信息与待嵌入水印的小波系数的比值与相对阈值进行比较的方法来选择水印嵌入位置更能体现出水印对待嵌入系数自身的影响程度，更具有自适应性。
英文文摘	With the rapid development of the digital multimedia technology and computer networks, it is easy to duplicate, modify and transmit the digital works, the copyright protection becomes a serious problem day by day. Digital watermarking is a new technology of copyright protection of digital multimedia, the research on it is still at the starting stage, such problems as the basic principle, key technology and application are needed to be study further. The correct and rational systematic frame is to the assurance of carrying out research on digital watermarking. Analyze the basic frame, study the whole characteristic of the frame and evaluation criterion, and analyze the function and technological characteristic of each component. Carry on analysis of the robustness, imperceptibility and security of the digital watermarking, do deep investigation on the methods to realize and strengthen these characteristics. Robustness and imperceptibility are contradicted innately, it is a difficult problem to realize giving consideration to the two. To carrying on deep researching and analyzing by utilizing the image content and vision characteristic on adaptive algorithm, described the certainty of its outcome, essence and advantages. Improve the adaptive algorithm based on discrete cosine transformation. Calculate the averages of coefficients of low frequency district and intermediate frequency district coefficients separately, choose the coefficients that lies between these two averages as the embedding positions; utilize the vision distance model to calculate the distance between the cover image and watermarked image after embedding, then compare this distance with the predefined threshold, if it is greater than threshold, adjust the embedding strength to embed again, in this way, guarantee the enough robustness and fully guarantee imperceptibility also. Improve the adaptive algorithm based on discrete wavelet transformation. Utilize the suitable vision model on wavelet domain and make use of the image content and vision characteristic to calculate the just noticeable differences of each wavelet coefficients, multiply these just noticeable differences with corresponding factor to be the embedding strength, it makes the strength more accurate; compare the ratio of watermark and the coefficient to be embedded and predefined threshold, choose the coefficients that are smaller than the threshold to as the embedding position, use the ration and the relative threshold to choose the embedding position can embody the influences that the watermark causes to the coefficient more efficiently, make the watermark more adaptive.