基于载体视频选择的信息隐藏技术研究

The rapid development of steganalysis technology brings serious challenges to the classical information hiding, especially on its security, which makes the research on new information hiding technology imperative. In this project, the video is selected as the steganographic cover, and the cover-selection based information hiding is set as the main content, where the cover is selected via the matching between video hash and steganographic information. The research is started from the video feature extraction, and will focus on improving the robustness of information hiding, the capacity of hidden information, and the efficiency of cover matching based on the optimization of video hash, which is on the basis of video feature extraction and fusion. We will extract the appearance, attention and memory features from the stimulus-perception-memory levels of visual perception system to generate the multi-level descriptor of the video, and use the multi-level descriptor as the information hiding domain. We aim to provide the robustness of information hiding through this way. We will fuse the multiple features through deep neural network, tensor decomposition, and independent component analysis, and improve the efficiency of cover matching based on feature fusion. We will take use of the temporal information of the video, and model the relationship among the length, robustness, and recall rate of video hash based on the theory of rate-distortion. We plan to improve the capacity of hidden information via optimizing the model. We try to provide the theoretical reference and technical support to the practical application of cover-selection based information hiding through the research of this project.

隐写分析技术的迅猛发展对经典信息隐藏技术的安全性提出了严峻的挑战，新型信息隐藏技术的研究势在必行。本项目以视频为隐写载体，以基于载体选择的信息隐藏为研究对象，以基于视频哈希与隐写信息匹配的载体视频选择为基本出发点，从视频特征提取和融合入手，通过优化视频哈希，改善信息隐藏鲁棒性、隐写信息容量以及载体匹配效率三个方面的性能。提取视频在视觉认知的刺激-感知-记忆层次上的表观特征、关注特征和可记忆性特征，以视频的多层次特征为信息隐藏域，从根本上确保信息隐藏的鲁棒性；采用深度神经网络、张量分解和独立分量分解等方式，通过视频特征的融合，优化信息隐藏域，为改善载体的匹配效率提供保障；借助视频的时域特性，以率失真优化为基础，通过对哈希码长-鲁棒性-查全率的建模，在保证性能的前提下，提升隐写信息的容量。本项目力争通过研究为基于载体选择的信息隐藏技术的实用提供理论参考和技术支持。

隐写分析技术的迅猛发展对基于载体修改的信息隐藏技术的安全性提出了严峻的挑战，本项目以安全性更高的基于载体选择的信息隐藏技术为研究目标开展研究。本项目从视觉特性建模和可记忆性预测、视频时空特征的提取、视觉特征与文本的映射、哈希映射和哈希码长的优化、基于载体选择的信息隐藏等五个方面开展了研究，获得了基于不同模式复杂度、结构规则性、图像块方向特征、颜色复杂度、视觉内容分类、关注焦点信息及其周边信息等JND模型；获得了基于三维空间卷积块和三维时间卷积块的视频特征提取方法；获得了基于特征选择优化、基于标注数据传播、基于子空间联合学习的视觉特征与文本之间的映射方法；获得了基于继承映射的视觉特征空间到哈希空间的映射方法，获得了基于类内误码率和类间误码率的概率分布建模的哈希码长优化方法；获得了基于子图像平均值的无载体信息隐藏方法，基于帧间组合的无载体视频信息隐藏方案。.本项目的研究在反映视觉敏感性的JND建模、视频特征提取和融合、视觉特征与文本特征的跨模态映射、哈希码长的优化、基于图像和视频载体选择的信息隐藏等多个方面取得了一系列的研究成果，有助于改善基于载体选择的信息隐藏的鲁棒性、信息容量以及载体匹配率等三方面的性能，为基于载体选择的信息隐藏技术的实用提供理论参考和技术支持。

内容获取失败，请点击重试