Theory and Algorithms for Robust Information Embedding

Massachusetts Institute Of Technology, Cambridge MA

Investigators

Abstract

A Research Project on Theory and Algorithms for Robust Information Embedding Principal Investigator: Prof. Gregory W. Wornell Research Laboratory of Electronics Massachusetts Institute of Technology ABSTRACT A 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.