可视媒体取证新技术研究

Recent development in digital multimedia processing methods and related editing softwaretools, such as Adobe Photoshop and Premiere,has made it increasingly easy for ordinary users to tamper the contents of multimedia data without leaving obvious visual traces. Nowaday, seeing is no longer believing. And it will get worse as counterfeiting techniques get more and more sophisticated. Authentication of multimedia data faces great challenges.In this project, we focus on several new techniques in the visual media (including digital image and video) forensics, and the innovation of the project is list as follows:..1. Most existing works just employ the early vision information in an image for the purpose of multimedia forensics. Based on the reported results, it is found that their performances are far from satisfactory, and they usually are sensitive to some post-operations. In this project, we try to propose an interactive forensic scheme to make full use of the mid-level and high-level vision information for further improving the related works on image splicing and light consistency detection...2. Seamless tampering (including object splicing and removal) is well studied in the research areas of computer vision and computer graphics. Some technologies ready and mature for application have already been included in the new release version of Photoshop CS5 in April 2010. However, the corresponding forensic techniques have not been proposed until now. In this project, we first try to analyze the statistical artifacts introduced by the seamless tampering both in digtial image and video...3. Typically, there are many available source encoders for digtial image and video. If the tampered image/video has been converted from one source encoder to another, all existing works on recompression detection will become useless due to lack of the re-quantization artifacts in the corresponding frequency domains.In this project, we try to expose the visual media after different encoder conversion, and then further to estimate some important parameters used in the primary encoder...4. Sensor pattern noise and Demosaicking algorithm used inside the digital camera will introduce an inherent fingerprint for the output images. Many forensic works use such a fingerprint to identify camera individual and/or camera model. However, just a few works have been considered in the view of anti-forensics. In this project, we try to propose an image content adaptive scheme to re-embed the fingerprint into images to fool the existing forensic methods.

随着各种媒体处理方法及其软件的发展, 普通用户都可很简单地实现对媒体篡改且不留下视觉上痕迹。如今，眼见并不为实，媒体取证分析正面临严峻挑战。本项目将针对可视媒体（图像与视频）取证中的若干新技术进行深入探讨，其主要创新点包括：1.目前大部分取证方法仅利用了图像的低层视觉信息。项目拟通过人的简单交互，充分结合图像中高层视觉信息，以提高现有算法的检测性能；2.无缝篡改在计算机视觉、图形学中已研究多年，一些较成熟算法亦已添加到Photoshop新版本中，但相应的取证技术却尚未见报道。项目将首次分析图像/视频的无缝篡改；3.现有重压缩检测技术无法判别经不同源编码器转换后的图像/视频，项目拟研究不同源编码器的转换检测及参数估计；4.基于模式噪声与颜色插值特征的相机源识别算法有许多，但从反取证角度出发的研究报告且极少。项目拟提出一种图像内容自适应的模式噪声、颜色插值重嵌入策略，以混淆现有相机源识别算法。

如今，数字媒体（主要研究数字图像及视频）已经成为人们获取及传递信息的最主要载体之一。然而，随着各种媒体处理工具的迅速发展，篡改且不留下任何视觉上痕迹变得越来越容易，这将不可避免地对我们个人隐私、法律取证、科学发展乃至国家安全带来一定的负面影响。数字媒体取证具有十分重要的研究及实际意义。本项目主要围绕着媒体安全若干关键问题展开研究，主要研究成果包括了：1. 基于图像残差统计特征的通用型取证分析；2. 图像Inpainting 的篡改检测技术; 3. 基于篡改概率图融合机制的篡改定位 ；4. 基于JPEG图像的取证与反取证技术；5. 自适应隐写分析方法及在取证中的应用探讨; 6. 数字音频压缩历史分析; 7. 视频原始码率、帧率识别等。