面向可视媒体内容安全的取证技术研究

项目名称： 面向可视媒体内容安全的取证技术研究

项目编号： No.61272355

项目类型： 面上项目

立项/批准年度： 2013

项目学科： 自动化技术、计算机技术

项目作者： 倪蓉蓉

作者单位： 北京交通大学

项目金额： 80万元

中文摘要： 可视媒体内容安全仍未解决的关键问题集中在内容真实性的验证上。涉及的核心技术体现在：基于水印的主动认证和自恢复、基于特征分析的被动取证。本课题针对图像和视频，研究可视媒体语义内容安全的新理论和新方法。研究内容包括：1）高准确性的主动认证与自恢复。根据图像的统计特性、内容一致性的关键特征，基于超像素提取对恶意篡改敏感的有效信息，认证内容真实性并且对失真内容进行修复。2）抵抗无恶意篡改的主动认证。当同时存在压缩、滤波等无恶意的处理操作时，仍能对篡改区域进行定位。篡改检测方案具有一定的鲁棒性，建立内容变化与频谱特性的内在关系，实现内容认证。3）图像拼接合成的取证。通过对图像内容的局部和全局特征分析，寻找具有普适意义的特征组合，设计结合启发式分类和集成学习的盲取证方案，结合错误学习机制，实现图像的真实性鉴别。4）视频拷贝粘贴的取证。针对视频片段，基于帧融合思想和相似性度量方法，准确检测拼接拷贝。

中文关键词： 可视媒体内容安全；数字水印；数字取证；生物认证；特征分析和分类

英文摘要： The remain unsolved key question of the multimedia content security relies on verification of the content integrity. The involved key technology depends on "watermarking based active authentication and self-recovery" and "passive forensics based on media feature analysis". This project mainly focuses on the research of new theories and methods of syntax content security combining human vision characteristics, pattern-classification and etc tools. The research scope focuses on: 1) Active authentication and self-recovery with high precision. This will base on the statistics feature and content consistency of digital image and use superpixels to extract the valid information sensitive to malicious tampering, then authenticate the integrity of content and recover the distorted content. 2) Active authentication of resisting content-preserving changes. This will provide the capability of locating the changed region after the compression, filter and other unmalicious operations to the content and establish the internal relationship between content variation and frequency to implement the content authentication. 3) Forensics of the image splicing. By analyzing the local and global image features, this method will research the universal feature combination and design certification system combing assembled learning and e

英文关键词： Multimedia Content Security；Digital Watermarking；Digital Forensics；Biometrics；Feature Analysis and Classification