基于数据驱动的图像隐私保护算法安全分析与设计

Using mobile internet as transmission media, images and videos are transmitted from the source of recording camera to the receivers with an amazing fast speed via social networks and cloud servers. They may contain a lot of privacy information relating to personal critical interests and even that of a country. The key strategy of a privacy protection scheme is to disturb the relationship between the sensitive information and a given natural person via encryption. This project aims to analyze security of some typical image encryption schemes and surveillance video privacy protection schemes based on position permutation and discrete transform, from the perspective of modern cryptanalysis..Concretely, the capabilities of the schemes against extraction of image privacy characteristics and encryption keys will be analyzed using the strategy of image data-driven and the method of machine learning. Both rigorous mathematical analyses and detailed experimental results will be provided to support the found points, which will quantitatively demonstrate the restraining relationship among size of encryption object, computational load of encryption, structure of privacy encryption scheme and security level against various attacks. Based on comprehensive survey of design and security analysis of image (video frames) privacy protection schemes, we will summarize the general guidelines for designing secure and efficient image privacy protection schemes. Finally, we will design a surveillance video privacy protection scheme adjusting the involved factors automatically according to the real environment and implement it as a real demo. The expected results of this project will provide novel technological support for securing network-application-oriented data and privacy protection.

以摄像头为源头、移动互联网为媒介，图像和视频通过社交网络和云服务器飞速传播。这些文件中携带的大量隐私信息可能涉及个人和国家的切身利益。隐私保护算法主要通过加密来破坏敏感信息与特定自然人之间的关系链。本项目从现代密码分析学的角度对一些基于位置置换和离散变换的典型图像加密算法和监控视频图像帧隐私保护算法进行安全分析。拟采取图像数据驱动策略和机器学习方法来分析这些算法抵抗各种条件下提取图像隐私特征和加密密钥的能力，并提供严谨的数学证明和详细的实验结果。这将定量地揭示加密数据量、加密计算复杂度、隐私保护算法结构和抗隐私提取安全水平等因素之间的约束关系。在全面分析总结图像隐私保护算法分析与设计工作的基础上，总结设计安全高效隐私保护算法的一般性法则。最后，我们将设计一个能自动调节各相关因素平衡的监控视频隐私保护算法，并提供示范系统。本项目的研究成果将为面向网络应用的数据安全机制和隐私保护提供技术支撑。

图像数据攫取、存储、传播设备的持续改进和社交网络的普及使得图像隐私保护成为大众关注的焦点。本项目采取图像数据驱动策略，从现代密码分析学的角度对一些典型图像加密算法和监控视频图像帧隐私保护算法进行安全分析。着重给出了8个算法抵抗唯密文攻击、已知明文攻击、选择明文攻击的能力。为全面揭示所用混沌系统生成伪随机序列的安全性能，解析了Logistic映射、Tent映射、Cat映射等经典映射的内在结构随计算机实现精度递增时的变化规律。设计了一个简化的多涡卷蔡氏电路，并用于伪随机数生成器。严谨地证明了所有相关理论分析，并得到实验验证。在密码分析和全面综述的基础上，提炼了设计安全高效图像加密和隐私保护算法的一般性法则。发表IEEE汇刊系列论文6篇、ESI高被引论文5篇、完成学术专著一部、获省自然科学二等奖一项、参与项目的4名硕士生继续攻读博士学位。本项目的研究成果将为面向网络应用的图像数据安全机制和隐私保护提供基础支撑。