JPEG压缩域图像增强取证研究

In recent years, the incidents that use tampered images to commit crimes occur quite frequently, which have done harm to the public and personal interest. Image enhancement forensics can detect and locate tampered regions by analyzing the abnormality of enhancement traces and becomes one of the important directions in image forensics. Currently, some progresses have been made in the field of image enhancement forensics, but all of them are targeted on pixel-domain enhancement, leaving its compressed-domain counterpart still unexplored. Directly applying the forensics methods designed for pixel-domain enhancement to solve the issues of compressed-domain enhancement forensics likely leads to poor performance. Therefore, it is in great need of conducting specialized research on the forensics of compressed-domain enhancement. This project is the first one to study the forensics of compressed-domain enhancement. The core idea is to construct the reverse operation of a compressed-domain enhancement algorithm, by which the problem of compressed-domain enhancement forensics can be transformed into its equivalent problem of JPEG compression forensics. The project mainly studies (1) the theoretical condition of reverse operation and its construction; (2) the likelihood modeling of enhancement parameter and the design of the methods of enhancement detection and parameter estimation; (3) the implementation of a prototype system for tampering detection and localization based on the consistency of enhancement traces. The output of this project not only provides a set of forensics tools for compressed-domain enhancement, but also provides a new route for the forensics of other manipulations in compressed-domain.

近年来，利用篡改图像进行违法犯罪的事件时有发生，对公共和个人利益造成了一定危害。图像增强取证通过分析增强痕迹的异常，能够实现篡改检测和定位，是图像取证的一个重要研究方向。目前，图像增强取证的研究取得了一定进展，但成果主要针对像素域增强，尚没有关于压缩域增强的取证工作报道。将像素域增强取证算法直接用于压缩域增强取证，通常效果很差甚至完全无效，因此有必要开展针对性的研究工作。本项目属于首次对压缩域增强进行取证研究，核心研究思路是通过构造增强算法的逆向操作，将增强取证问题转化为压缩取证问题进行分析和求解。本项目主要研究内容包括（1）逆向操作存在的理论条件和构造方法探索；（2）增强参数似然模型的构建，以及增强检测及参数估计算法的设计；（3）基于增强痕迹一致性的图像篡改检测和定位原型系统的实现。本项目的研究成果不但具有直接的应用价值，还将为其它压缩域操作的取证提供新的思路。

本项目围绕JPEG压缩域增强取证，研究了JPEG压缩和压缩域增强的作用机理，设计了JPEG压缩参数估计性能提升框架和压缩域增强取证算法，克服了现有方法对小尺寸图块估计精度差、对压缩域增强混叠痕迹取证能力弱的不足，显著提升了对小尺寸图块、压缩增强痕迹相互混叠情况的取证适用性和鲁棒性。主要创新和贡献包括：..1）提出了基于层次聚类的量化步长估计性能提升框架。基本思想是通过层次聚类汇集多个频率的系数样本，从而缓解单个频率系数样本不足造成的估计精度下降。本文以系数方图和因数方图作为聚类特征实现了所提框架，实验结果表明所提框架能够显著提升现有步长估计算法在小尺寸图块上的估计精度。..2）提出了基于似然建模的压缩域图像增强取证算法。基本思想是通过推导去增强操作，将压缩域增强造成的混叠痕迹转化为参数化压缩痕迹，后者更易于建模和求解。基于这一思想，本文构造了去增强系数的似然函数，并根据似然函数的性质为两种经典的压缩域增强设计了取证算法，能够同时实现增强检测、增强参数估计和压缩参数估计。大量对比实验验证了所提算法的有效性和性能优势。..在小尺寸图块上的取证精度低和多种操作痕迹相互混叠导致的取证困难，不只是JPEG压缩取证面临的挑战，也是其他图像取证任务普遍遇到的挑战。本项目通过两项工作展示了如何应对这些挑战，对其它图像取证任务应有一些启发和帮助。..本项目研究成果已整理发表3篇SCI权威期刊论文，其中2篇发表于国际多媒体信号处理核心期刊IEEE Transactions on Circuits and Systems for Video Technology；已申报中国发明专利4项。项目执行期间，课题组获得了媒体取证领域省市级项目的进一步资助；团队中2名成员顺利取得博士学位，2名硕士生顺利取得硕士学位，各成员的科研素养和技术能力得到显著提升。