面向多媒体密域的客户端数字指纹同态化嵌入理论和技术研究

Encryption and watermarking are both adopted to secure the digital multimedia product disseminated from the sever to the client. It is easy to encrypt the watermarked data while it's not easy to embed watermarks into the encrypted data in some untrusted environments. Till now the problem of singal procesings in encyption domain has been attracted more intentions including the encryption domain watermarking due to its great importance to the privacy, security and trust of many applications in networks. However in multimedia security applications most of them are absent of perceptual theory model and have constraints within the algorithm design. In this proposal we aim at above problems and study a new Homomorphism based Client-side Fingerprinting in multimedia encryption domain. It includes that encryption domain fingerprinting system theory model combined with the perceptual coding, self-homomorphism theory and techniques of transfering watermarking domain and encryption domain with each other under the perceptual constraints, multimedia homomorphic encryption and full homomorphic encryption algorithm and the encryption domain fingerprint embedding method. Our purpose is to disclose the perceptual theory rule of every performance factor influencing each other in the fingerprinting sysytem according to different selected encryption domain features, the realization condition of self-homomorphism transfering watermarking and encryption operation with each other, and finally to realize an efficient self-homomorphism or homomorphism encryption or full homomorphism encryption domain fingerprinting, which will provide the theory bases and efficient technique supports to the solution of the client-side fingerprint secure embedding problem in multimedia digital secure broadcast applications. And it also has the inspiration and significance to promote solutions of other problems in signal processing in encryption domain.

多媒体安全分发通常同时采用加密和水印技术。面对水印对象加密易实现，而在非安全环境下对密文嵌入水印却是难点问题。由于对安全隐私保护应用具有重要价值，包括密域水印在内的密域信号处理研究得以广泛关注。然而，当前多媒体密域水印缺乏感知理论模型指导而且算法构造存在束缚。为此本项目针对面向多媒体加密对象的客户端数字指纹同态化嵌入理论和技术展开研究。具体内容包括：研究感知编码条件下密域指纹系统理论模型；研究水印域与密域在感知条件下的互同态理论和技术；研究有效的多媒体(全)同态加密算法和密域指纹嵌入方法。本项目旨在揭示在感知理论条件下同态化密域特征对指纹系统各性能指标影响规律；水印操作和加密操作的互同态实现条件；并实现感知编码条件下有效的互同态密域指纹方案或基于（全）同态加密的指纹嵌入方法，为解决多媒体安全广播中的客户端指纹安全嵌入问题提供理论基础和技术支撑,并对密域信号处理研究的其它问题有重要启发意义。

互联网大数据和云计算快速发展的背景下，面向多媒体加密域的信号处理理论和方法为安全水印嵌入、隐私保护识别和检索以及安全压缩转码应用提供了技术途径。本项目针对以上应用分别在非安全环境下的同态化密文水印嵌入、多媒体加密域检索、安全可分级转码分发以及云端属性加密等方面进行研究，取得了一系列成果，可为多媒体安全分发与隐私保护提供关键支撑。重要结果包括：1)基于一种结合水印的多媒体加解密方法实现广播条件下的安全指纹嵌入算法，以及面向加密域图像提出一种大容量的可逆水印算法，为多媒体加密域的同态化指纹嵌入提供了更有效方法。2）面向大规模图像环境下的安全检索问题，基于部分加密思想实现互同态条件下的隐私保护生物特征图像检索方法和系统，相比明文检索效率和准确性接近；并在多特征融合的基础上通过哈希学习，得到效率效果更优的大规模图像检索方法。3）针对多媒体指纹分发应用的隐私保护和可分级要求，在云环境下外包解密的基于属性加密研究方面提出了一种可验证的密文策略的基于属性加密改进方案，并在此基础上提出一种更优的H.264/SVC的可分级视频安全转码分发方案。 .另外，课题组还对安全媒体指纹生成和鉴定的方法、语义安全的云端隐私保护关联规则挖掘、多媒体隐写分析与取证、群智感知安全等方面进行了研究并有重要结果，是对本项目云环境下多媒体加密域信号处理研究的衍生工作，具有积极的支撑意义。.课题组已发表学术论文26篇，其中被SCI收录14篇，EI收录21篇。有5项国家发明专利公开，其中1项已授权。指导硕士生4名，指导博士生3名。参与主办学术会议3次，其中承办多媒体安全重要学术会议CIHW2016。