Theory and Algorithms for Robust Information Embedding

鲁棒信息嵌入理论与算法研究项目摘要数字计算和通信广泛传播的一个后果是，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，从而变得越来越容易和灵活。这些功能开辟了一系列广泛的新应用，从照片和录音的数字修复，到通过网络方便、即时地分享这些内容。与此同时，这些变化具有深远的社会影响，使个人更容易绕过版权法，或绕过基于照片身份证的安全系统，或在法律审判中医生录音证据。信息嵌入是一种重要的新兴信号处理技术，用于管理此类多媒体内容问题。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多媒体“主机”信号或图像中，从而使嵌入既有效地难以察觉，又能抵抗良性形式的扰动和噪声的影响，也能抵抗明确的删除或修改嵌入信息的企图。虽然各种各样的启发式嵌入方法已经在研究界被引入，但这些现有的方法有许多缺点。这个研究项目正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，用于广泛的应用。这些应用包括数字多媒体内容的版权保护、广播的自动监控、文件的认证和身份识别。研究的重点是基于量化的鲁棒嵌入方法，这些方法使它们能够很好地用于实际实现。作为一个特别的研究重点，正在开发一类称为量化指数调制（QIM）的方法。部分研究涉及QIM的优化实现的开发，以及相对于替代方法的性能特征的评估。

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.