图像取证鲁棒性和安全性研究

With the rapid development of digital image editing techniques, image tampering and illegally alteration occur frequently. A huge hidden trouble has been brought to the construction of social honesty and judicial justice. As such, the image forensics technique is urgently needed. Currently, robustness and security are the two major unsolved problems in image forensics techniques, where the robustness refers to capabilities resisting post-processing and adapting to pre-processing, and the security refers to the capability resisting anti-forensics attacks. This project will make a study of the theory and methods of such robustness and security. For such purpose, the scientific theories including pattern recognition, information fusion and statistical signal processing, and the prior knowledge on the statistical characteristics of both imaging pipeline and natural images are utilized. Specifically, we will primarily study the following subjects: the robustness evaluation methods based on testing and proving, the image forensics techniques based on information fusion, the image manipulation history forensics, and the theory and methods of security evaluation based on hypothesis testing. The research achievements of this project can be extensively applied to the fields including judicial forensics, criminal evidence and news photos supervision, and have great significance in both theory and practice.

随着数字图像编辑技术的快速发展，图像篡改或违规编辑现象频繁发生，给社会诚信建设和司法公正带来极大隐患，图像取证成为一项迫切需求的技术。当前，图像取证技术中仍未解决的关键问题集中在鲁棒性和安全性上。鲁棒性指抵抗后处理和适应预处理的能力，安全性指抵抗反取证攻击的能力。本课题拟借助模式分类、信息融合和统计信号检测等科学理论，结合成像通道与自然图像的统计特性，研究图像取证鲁棒性和安全性的理论与方法，主要包括：基于测试与论证的鲁棒性评估方法，基于信息融合的图像取证技术，图像操作历史取证，以及基于假设检验的安全性评估理论与方法等。本项目的研究成果可以广泛应用于司法取证、刑事取证和新闻照片监管等领域，具有重要的理论意义和实用价值。

随着数字图像编辑技术与工具的应用普及，对图像数据与内容的修改变得更容易且不留痕。在实际应用中，恶意用户常在未经授权的情形下对数字图像进行非法操作，如违规编辑、合成虚假图像、篡改可视化证据等，对证据鉴别、新闻纪实、司法取证、刑事取证等图像信息安全应用构成重大威胁。因而，研究可信的数字图像取证技术具有十分重要的理论意义和应用价值。本项目主要围绕图像取证的安全性和鲁棒性理论与方法开展研究，取得的主要研究成果包括：1）提出了基于假设检验的图像操作取证安全模型，并给出基于重采样操作痕迹伪造型攻击的应用示范；2）对先进图像对比度增强算法，及其取证与反取证方法的深入分析与设计；3）对几何变换鲁棒的图像统计矩特征设计；4）对桶形几何失真鲁棒的视频主动取证算法设计；5）基于人脸表情分析的感知取证方法设计。