隐写技术对于保护民众隐私和传递机密信息都具有重大的意义和应用价值。现有基于单一特征的隐写方法存在隐藏容量和隐蔽性的相互制约关系，难以满足用户高隐藏容量、高隐蔽性的隐写需求。网络载体具有数据规模大、层次结构多、模态种类多、载体时序化、媒体相关联等特点，使得网络环境下多维时空隐写方法具有更大的隐藏容量和隐蔽性优势。本文关于网络环境下多维时空隐写方法，开展了以下四个方面的研究： 为了增强时序载体隐写模式的适应性，提出了一种多特征隐写方法。该方法通过建立网络时序载体的多维时空隐写框架，使用正交的隐写特征分散嵌入秘密信息，结合多种隐写策略，组成灵活多样的隐写模式。实验结果验证了该方法可以有效地增强抗隐写检测的能力来适应高隐蔽性要求的通信环境。 为了增强异构载体的语义相关性，提出了一种基于语义相关性的多模态隐写方法。该方法利用异构载体语义相关性的度量模型，学习网络空间中多模态数据的特征表示，将秘密信息编码嵌入到特征表示相似的异构载体的语义相关性中。实验结果验证了该方法可以大幅度提高隐写载体的语义相关性。 为了增强同构载体的上下文连贯性，提出了一种主题约束的载体生成式隐写方法。该方法利用循环的层次化主题模型，学习上下文的主题分布，在主题分布的约束下生成含密载体。实验结果验证了该方法可以生成上下文连贯性更高的含密载体。 研究设计了一个面向公共社交网络环境的多维时空隐私保护系统。该系统利用正交的隐写特征，将秘密信息嵌入到多模态隐写载体中；通过构建隐写矢量能力集，实现收发方的同步；使用差错控制和自检错误重传机制保障信息在非受控网络环境下传输的正确性。测试结果验证了该系统在功能实现上的有效性。
Steganography technology is significant and promising for privacy protection and secret message transmission. Since current steganographic methods are based on single hiding feature, their hiding capacity and imperceptibility are mutually restrictive. Therefore, current steganographic methods cannot meet the high-capacity and high-imperceptibility requirements. There are an enormous amount of sequential multi-modal covers under network environment. These covers have multi-layer structure and relevancy to each other. Thus multi-dimensional spatial-temporal steganography has higher hiding capacity and imperceptibility under network environment. This dissertation engages in the research of multi-dimensional spatial-temporal steganography. The main contributions of this dissertation are summarized as below. A multi-feature based steganographic method is proposed to enhance the adaptability of the hiding mode on sequential covers. This method utilizes orthogonal hiding features to embed the secret messages separately by establishing a multi-dimensional spatial-temporal steganographic framework. Multiple hiding strateges are combined to make the hiding mode flexible. Experiment results demonstrate that this method is effective to enhance the resistance to the steganalysis methods, thus being adapted to the communication environments which have the high-imperceptibility requirement.A multi-modal steganographic method based on semantic relevance is proposed to enhance the semantic relevance of the heterogeneous covers. This method utilizes a model to measure the semantic relevance of the heterogeneous covers and learn the embeddings of the multi-modal data. Secret messages are embedded into the semantic relevance of the heterogeneous covers whose embeddings are similar. Experiment results demonstrate that this method can improve the semantic relevance of the hiding covers to a large extent.A topic-constrained steganographic method based on cover synthesis is proposed to enhance the contextual coherence of the homogeneous covers. This method utilizes a recurrent hierarchical topic model to learn the topic distribution of the context. The stego covers are synthesized under the constrains of this topic distribution. Experiment results demonstrate that this method can synthesize stego covers with higher contextual coherence.A multi-dimensional spatial-temporal privacy transmission system is designed for the public social network. This system utilizes orthogonal hiding features to embed secret messages into the multi-modal covers. A hiding vector capacity set is constructed to synchronize the receiver and the sender. The completeness of the secret messages transmitted under the uncontrolled network environment is guaranteed by the error control and retransmission scheme. Test results demonstrate that this system can achieve the required functions.
