CASIA OpenIR  > 毕业生  > 博士学位论文
几何攻击鲁棒的视频水印算法研究
吕忠泽
2022-05-26
页数146
学位类型博士
中文摘要

随着信息和通信技术的高速发展,数字视频内容可以在网络上随意地被传播和使用,侵权现象时有发生,对其进行版权保护迫在眉睫。视频水印技术为数字视频内容的版权追踪与查证提供了解决方案,它将一个代表版权信息的标识(水印)不可察觉地嵌入到视频中,当遇到版权纠纷时,对视频中的水印进行提取来确定其所有权。

嵌入水印之后的视频在传播和使用中可能会遭受各种有意或无意的攻击,使其中的水印难以被正确提取。常见的攻击类型有几何攻击、信号处理攻击以及时间去同步攻击。在这几种类型的攻击中,几何攻击最难以抵抗,因为其破坏了水印在嵌入和提取时所在位置的同步性。其中,缩放和剪切是数字视频内容最常见的两种几何攻击,但目前的水印算法都难以同时有效抵抗这两种攻击。因此,研究设计几何攻击鲁棒的视频水印算法极具挑战性。

对于视频水印算法,视频帧的选择、嵌入区域的确定以及水印嵌入和提取方法是影响水印抵抗几何攻击的能力的关键因素。通常更倾向于选择部分帧进行水印嵌入以提升水印的不可见性和嵌入效率,但在提取时存在干扰帧降低了水印提取的准确率和效率;嵌入区域分为全局和局部,全局水印算法难以抵抗剪切攻击,而局部水印算法存在嵌入区域的定位误差,降低了水印提取的准确率;量化方法是目前最常见的一种水印嵌入和提取方法,其执行效率高且可以实现盲提取,但难以抵抗缩放攻击。

本文旨在缓解上述问题,从而设计几何攻击鲁棒的视频水印算法。本文主要工作和创新点归纳如下:

1. 为了提高水印抵抗缩放等多种几何攻击的能力,提出一种自适应量化索引调制视频水印算法。首先,基于直流(DirectCurrent,DC)系数的缩放不变性, 提出一种自适应量化索引调制方法,通过设置量化步长的大小随视频尺寸的不同而自适应地调整,实现了量化值和量化区间在缩放前后的同步性;其次,提出一种提取成功预先判别的策略,将快速响应码(QuickResponseCode,QRCode) 被解码成功作为水印提取成功的判别依据,减少了提取过程的执行次数;最后,基于QRCode的编码不变性,优化了水印的有效载荷。实验表明,该算法可以有效抵抗缩放等多种几何攻击及其他类型的攻击,且执行效率高。

2. 为了提高水印抵抗剪切等多种几何攻击的能力,提出一种基于双几何同步的视频水印算法。首先,提出一种双几何同步机制,通过引入模板辅助特征点定位嵌入区域,减小了特征点偏移或消失对嵌入区域定位所产生的误差;然后,提出一种基于场景切换的帧分配策略,将模板和水印分别嵌入到场景切换帧及其之后的若干帧中,避免了模板和水印的互干扰,减少了每一帧的数据修改量。实验表明,本文所提出的创新点对算法的性能提升都起到了积极作用,该算法可以有效抵抗剪切等多种几何攻击以及其他各种类型的攻击。

3. 为了进一步提高水印抵抗几何攻击的能力,提出一种基于帧序列的双重视频水印算法。首先,提出一种双重水印构建机制,将完整的水印嵌入到多个视频帧的全局区域中作为第一重水印,然后将水印按字节分段并将每一段分别嵌入到不同视频帧的局部区域中作为第二重水印,使双重水印分别抵抗不同类型的几何攻击。针对第二重水印,提出一种字节分段嵌入机制,利用帧序列的冗余性和里所码的编码优势,极大程度地减少了有效载荷。同时,提出一种定位误差补偿策略,通过对DC特征向量进行差值量化,来抵消定位误差引起嵌入区域所遭受到的小尺度偏移。实验表明,该算法对缩放和剪切等多种类型的攻击都具备更优秀的抵抗能力,综合性能更强。
 

英文摘要

Digital video content can be freely utilized and distributed on the Internet, thanks to the rapid advancement of information and communication technology. Infringement does happen from time to time, and it is critical to protect the copyright of digital video content. Video watermarking technology provides a solution for copyright tracking and verification of the digital video content. It embeds a logo (watermark) representing the copyright information imperceptibly into the video, and extracts the watermark from the video to determine its ownership when encountering the copyright disputes. 

The watermarked video may be vulnerable to a variety of planned and unintentional attacks during transmission and usage, making accurate extraction difficult. Geometric attacks, signal processing attacks and time desynchronization attacks are  the common types of attacks. Among them, geometric attacks are the most difficult attacks to resist, as they disrupt the watermark location's synchronization during embedding and extraction. Furthermore, scaling and cropping are the two most common geometric attacks on digital video content, but current watermarking algorithms have weak resistance to both attacks at the same time. As a result, researching and designing geometric attacks invariant video watermarking is extremely challenging.

The selection of video frames, the determination of embedding regions and the watermark embedding and extraction methods are the key factors, which affect the resistance of the watermark to geometric attacks in video watermarking. In order to improve the imperceptibility and the execution efficiency,  it is common practice to select the partial frames for watermark embedding . However, interference frames in the extraction process reduce the accuracy and efficiency of watermark extraction. There are two types of embedding regions: global and local. Cropping attacks are difficult to resist with global watermarking, whereas location errors in local watermarking reduce the accuracy of watermark extraction. Quantization is currently one of the most popular watermarking embedding and extraction methods, with high computational efficiency and the ability to perform blind extraction, but it is vulnerable to scaling attacks.

The goal of this dissertation is to solve the problems mentioned above, by designing the geometric attacks invariant video watermarking algorithms. The main contributions and innovations are summarized as follow:

1. In order to enhance the resistance of the watermark to various geometric attacks such as scaling, an adaptive quantization index modulation video watermarking is proposed. Firstly, based on the scaling invariance of the direct current(DC) coefficient, an adaptive quantization index modulation watermark is presented. To achieve the synchronization between the quantization values and the quantization intervals before and after scaling, it modifies the sizes of the quantization steps adaptively according to the sizes of the distinct videos. Following that, a strategy to anticipate the success of the extraction is proposed. It employs the successful decoding of the Quick Response Code(QRCode) as the criterion for judging the success of watermark extraction. It decreases the execution iterations and improves the efficiency and accuracy of watermark extraction. Meanwhile, the payload of the watermark is optimized to improve the performance of the algorithm, due to the encoding invariance of the QRCode. Experiments demonstrate that the algorithm can effectively withstand many types of attacks such as scaling, while maintaining a high level of execution efficiency.

2. In order to improve the ability of the watermark to resist various geometric attacks such as cropping, a video watermarking based on double geometric synchronization is proposed. To begin, a double geometric synchronization mechanism is presented. It presents the template to aid the feature points in locating the embedding regions, decreasing the errors caused by the feature points offset or disappearance. Then, based on scene change, a frame allocation strategy  is given. The template and the watermark are  respectively embedded in the scene change frames and several subsequent frames by this mechanism. It avoids the mutual interference between the template and the watermark and decreases the  amount of data modification in each frame. Experiments show that the above innovations have improved the performance of the algorithm. The algorithm is capable of resisting a variety of attacks, including cropping.

3. In order to further enhance the robustness of the watermark against geometric attacks, a dual video watermarking based on frame sequences is proposed. First of all, a mechanism for constructing dual watermarks is proposed. As the first watermark, a complete watermark is embedded into multiple complete frames. As the second watermark, the watermark is segmented by bytes and each segment is respectively embedded into the local regions in distinct frames. Therefore, the dual watermarks can separately resist multiple forms of geometric attacks. For the second watermark, a byte segmentation embedding mechanism is presented. It takes advantage of the frame sequences' redundancy and the Reed-Solomon Code's encoding advantages  to drastically decrease the payload. Meanwhile, a location errors compensation strategy is proposed. The small-scale migrations on the embedding regions induced by the location errors are mitigated, by carrying out differential quantization on the DC feature vector. Experiments demonstrate that the algorithm is more resistant to many forms of attacks, such as scaling and cropping, with a better overall performance.

关键词视频水印 几何攻击 量化 模板 双重水印
语种中文
文献类型学位论文
条目标识符http://ir.ia.ac.cn/handle/173211/48461
专题毕业生_博士学位论文
推荐引用方式
GB/T 7714
吕忠泽. 几何攻击鲁棒的视频水印算法研究[D]. 中国科学院自动化研究所. 中国科学院自动化研究所,2022.
条目包含的文件
文件名称/大小 文献类型 版本类型 开放类型 使用许可
几何攻击鲁棒的视频水印算法研究_吕忠泽.(3103KB)学位论文 限制开放CC BY-NC-SA
个性服务
推荐该条目
保存到收藏夹
查看访问统计
导出为Endnote文件
谷歌学术
谷歌学术中相似的文章
[吕忠泽]的文章
百度学术
百度学术中相似的文章
[吕忠泽]的文章
必应学术
必应学术中相似的文章
[吕忠泽]的文章
相关权益政策
暂无数据
收藏/分享
所有评论 (0)
暂无评论
 

除非特别说明,本系统中所有内容都受版权保护,并保留所有权利。