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。
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.
|Keyword||视频水印 几何攻击 量化 模板 双重水印|
|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.|
|Files in This Item:|
|Recommend this item|
|Export to Endnote|
|Similar articles in Google Scholar|
|Similar articles in Baidu academic|
|Similar articles in Bing Scholar|
Items in the repository are protected by copyright, with all rights reserved, unless otherwise indicated.