加密域图像高性能压缩技术研究

Since the last decade, tremendous progresses in social networking and cloud computing have been achieved. These highly user-oriented applications generate large number of privacy-sensitive, user-related data, which unfortunately can be very easily accessed through communication networks, especially the Internet. Important concerns regarding the date security and privacy (especially after the Snowden affair) have thus been raised, calling for an appropriate protection mechanism without sacrificing the efficiency of authorized operations performed on them. .A standard way of protecting data from unauthorized access and manipulations is to apply cryptographic primitives such that only those with the secret key can disclose the data. However, in many practical scenarios, e.g., cloud computing, the parties who process the data might be untrusted and have no access to the secret key. This has led to the challenging problem of how to simultaneously achieve high level of security/privacy and superior efficiency of processing the encrypted data. In this project, we plan to develop high-performance compression technology for encrypted images. Specifically, we will investigate: 1) joint design of image encryption and image compression schemes such that the encrypted image can still be highly compressible, while still ensuring high level of security. 2) scalable coding of stream cipher encrypted images. The developed tools and platforms will benefit a rich variety of applications when dealing with privacy-sensitive multimedia data. .

斯诺登棱镜门事件曝光后，用户数据，特别是多媒体数据，如图像、视频、音频，的安全和隐私问题受到极大关注。一种简单而行之有效的保护信息安全和隐私的方法是数据加密。然而，这也给如何高效地处理数据带来新的挑战。现有的信号处理理论均基于信号的结构特性，并不适用于结构信息丢失的加密信号。在本课题中，我们研究加密图像的高性能压缩技术。围绕如下两个内容进行具有前瞻性的深入研究。首先，同时从加密和压缩入手，联合优化和设计一套加密和压缩算法，力图取得和传统非加密域压缩系统非常近似的压缩效率。其次，针对被经典加密算法保护的图像，设计一套高效的可伸缩编码系统。本项目的创新成果，对构建具有我国自主知识产权的多媒体安全平台，具有重要的理论意义和广泛的应用价值。

本项目以加密域图像高性能压缩和处理为核心，在三个相关领域开展了较为深入和系统的研究。在加密域图像的可伸缩编码领域，设计了一套基层编码和增强层编码。并提出了加密域自适应采样的机制，在总体采样点固定的前提下，大幅提高恢复图像的品质。在加密域图像数字水印领域，提出一种基于密钥调制的水印嵌入算法，将不同的水印映射到不同的密钥，利用异或操作，快速地嵌入水印信息。并利用图像的自相似属性，将水印提取问题转换成区分加密图像块和自然图像块问题。在高鲁棒性加密图像处理方面，研究了社交网络平台对图像的有损处理机制，有针对性地设计了一套DCT域的加密算法，使加密图像在经过有损处理后，仍能高质量地重构。. 项目共发表论文31篇，包括SCI期刊论文20篇，其中IEEE/ACM会刊论文12篇。