Theory and Algorithms for Robust Information Embedding

鲁棒信息嵌入的理论与算法研究项目主要研究者：Gregory W.麻省理工学院电子学研究实验室摘要数字计算和通信的广泛传播的结果是电子格式的多媒体内容-如音频、视频和静止图像-可以被分发、交换、操纵和以其他方式处理的越来越容易和灵活。这些功能开辟了一系列新的应用，从照片和录音的数字恢复到通过网络方便和即时共享这些内容。与此同时，这些变化具有深远的社会影响，使个人更容易，例如，在法律的审判中规避版权法、或规避基于照片ID的安全系统、或医生语音记录证据。信息嵌入是指以数字签名或指纹或其他比特序列的形式将信息嵌入到多媒体“主”信号或图像中，嵌入的方式既能有效地使人感觉不到，又能抵抗良性形式的扰动和噪声的影响，也能抵抗删除或修改嵌入信息的明显企图。虽然各种启发式的嵌入方法已被引入研究界，这些现有的方法有许多缺点。本研究项目旨在调查信息嵌入系统的基本限制，并探索接近这些限制的新方法，以实现广泛的应用。这些应用范围从数字多媒体内容的版权保护，广播的自动监控，文件的认证和identifiation.An研究的重点是鲁棒的基于量化的嵌入方法，这很好地借给自己的实际实现。作为一个特别的焦点，研究正在开发一类这样的方法，称为量化索引调制（QIM）。部分研究涉及QIM的优化实现的发展，并评估其性能特征相对于替代方法。

A Research Project on Theory and Algorithms for Robust Information EmbeddingPrincipal Investigator:Prof. Gregory W. WornellResearch Laboratory of ElectronicsMassachusetts Institute of TechnologyABSTRACTA consequence of the widespread dissemination of digital computing and communication is the increasing ease and flexibility with which multimedia content in electronic formats---such as audio, video, and still imagery---can be distributed, exchanged, manipulated and otherwise processed. These capabilities open up a broad array of new applications, from digital restoration of photographs and recordings, to convenient and instantaneous sharing of such content via networks. At the same time, these changes have far-reaching societal implications, making it easier for individuals to, e.g., circumvent copyright laws, or circumvent security systems based on photo IDs, or doctor voice recording evidence in legal trials.Information embedding is an important emerging signal processing technology for managing such multimedia content issues. Information embedding refers to the embedding of information in the form of a digital signature or fingerprint or other sequence of bits into a multimedia ``host'' signal or image in such a way that the embedding is both effectively imperceptible and resistant to the effects of benign forms of perturbation and noise, as well as to explicit attempts to remove or modify the embedded information. While a variety of heuristic approaches to embedding have been introduced within the research community, these existing approaches have many shortcomings. This research project is investigating the fundamental limits of information embedding systems, and exploring new methods for approaching these limits, for a wide range of applications. These applications range from copyright protection of digital multimedia content, to automated monitoring of broadcasting, to authentication of documents and identification.An emphasis in the research is on robust quantization-based embedding methods, which lend themselves well to practical implementations. And as one particular focus, the research is developing a class of such methods referred to as quantization index modulation (QIM). Part of the research involves the development of optimized implementations of QIM, and an evaluation of its performance characteristics relative to alternative approaches.