Cyber Trust - Restore the Trustworthiness of Digital Photographs: Blind Detection of Digital Photograph Tampering

网络信任 - 恢复数码照片的可信度：数码照片篡改的盲检测

• 批准号: 0430258
• 负责人: Shih-Fu Chang
• 金额: $ 74万
• 依托单位: Columbia University
• 依托单位国家: 美国
• 项目类别: Continuing Grant
• 财政年份: 2004
• 资助国家: 美国
• 起止时间: 2004-09-15 至 2008-08-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=0430258&HistoricalAwards=false
• 关键词: Cyber; Trust; Restore; Trustworthiness; Digital

Trustworthy photographs play an important role in many applications such as news reporting, intelligence information gathering, criminal investigation, security surveillance, as well as health care. However, with the advent of digital age, the trustworthiness of pictures could no longer be taken for granted. This project will develop a completely blind and passive system for detecting digital photograph tampering. No extra encryption, signature extraction, or information embedding processes are needed. Content tampering operations are detected at the point of checking by analyzing the natural signal/scene characteristics in the image. We take an innovative approach integrating techniques from signal processing and computer graphics. The signal processing method involves effective use of higher-order signal statistics, signal acquisition device modeling, image decomposition, and image structural analysis to identify tampering artifacts at the signal level. The computer graphics approach includes novel techniques of 3D geometry estimation, illumination field recovery, and scene reconstruction to detect inconsistency at the scene level like shadows, shading, and geometry. We aim at a successful system that makes any attacking maneuver as difficult as possible. The system will also provide equal emphasis on robustness and informativeness - suspicions will be explained with locations and reasons. To achieve broader impacts, the proposed research will include deployment of a public image forgery detection engine, release of a large original data set, and definitions of evaluation benchmark.

值得信赖的照片在许多应用中起着重要作用，例如新闻报道，情报信息收集，刑事调查，安全监视以及医疗保健。但是，随着数字时代的出现，图片的可信度不再是理所当然的。该项目将开发一个完全盲目的被动系统，用于检测数字照片篡改。不需要额外的加密，签名提取或信息嵌入过程。通过分析图像中的自然信号/场景特征，在检查时检测到内容篡改操作。我们采用一种创新的方法来整合信号处理和计算机图形的技术。信号处理方法涉及有效使用高阶信号统计，信号采集设备建模，图像分解和图像结构分析，以识别信号级别的篡改工件。计算机图形方法包括3D几何估计的新技术，照明场恢复和场景重建，以在场景级别（如阴影，阴影和几何形状）上检测不一致。我们的目标是成功的系统，该系统使任何进攻动作都尽可能困难。该系统还将同等地强调鲁棒性和信息性 - 将以位置和原因来解释怀疑。为了实现更广泛的影响，拟议的研究将包括部署公共形象伪造检测引擎，发布大型原始数据集以及评估基准的定义。