Boncelet's Research into
Steganography, Cryptography, and Security
Other Research:
In 1997, I gave the inaugural talk in ARL's Distinguished Lecturer series. You can download the powerpoint file entitled Digital Steganography and Watermarking . In October, 2001, I testified as an expert witness at the Science and the Law Conference sponsored by the National Institute of Justice.
Some specific research areas are the following:
Steganalysis
In January 2005, I received a grant from NSF to develop methods for staganalysis, the search for hidden messages in ordinary looking images, videos, audio files, and others. We are developing compression based steganalysis methods. Our first goal is to develop methods for still images. After all, there are millions of images out there on the web.
It is our contention that the best image modelers have come out of the image compression community. In the past decade or so, the compression community has been engaged in a brutal winner-take-all competition to build the best compressor. What does this have to do with modeling, you ask? Easy, lossless compression consists of two main parts: a modeler and an entropy coder. Typically the difference between the modeler output and the original data is encoded by the entropy coder. Entropy coders are lower bounded by Shannon's limits. Hence, better modeling means better compression.
In the summer 2005, we were awarded with a CIA postdoctoral fellowship.
Currently, I am working on JPEG steganalysis on funding from the US Army.
Steganography
Steganography, literally "covered writing" in Greek, refers to the hiding of a message within another, generally inocuous, message. Steganography should be contrasted with cryptography ("secret writing"). Both are used in security and both may be used together. Steganography is used primarily when the fact that you are communicating needs to be kept secret; cryptography is used to hide the meaning of your message, not the fact that you are communicating.In the last 1990's and early 2000's, I worked with ARL on steganography and developing new steganographic techniques. As one can imagine, the Army has both offensive and defensive interests in secret communications.
In the aftermath of the Sept. 11 terrorist attacks, I was interviewed by several publications who were researching the possibility that the terrorists may have used steganographic techniques to communicate. I have no knowledge one way or the other whether the terrorists actually did use steganography, but do know that it could have been done.
I will continue to work with the US government on these issues as they request, but will redirect my energies from creating new steganographic methods to attacking existing methods.
Some highlights of this work include the following:
- Spread Spectrum Image Steganography. Lisa Marvel of ARL and I have developed this technique to hide information in images and other multimedia objects. We have also studied the information theoretic capacity of additive steganographic systems.
- Stego-JPEG, a technique for hiding information in images. This technique is very good, both secret and practical. We have not published this work.
- I am currently working on a technique that hides information in random noise.
Message Authentication
I have developed several methods for Message Authentication in noisy environments. The first of these, the NTMAC (Noise Tolerant Message Authentication Code), works by breaking an object into multiple, overlapping pieces, computing a truncated conventional message authentication code (MAC) on each piece, and concatenating the truncated MAC's together. The receiver authenticates each piece separately. Any errors can be confined to the intersection of the pieces that fail the authentication.The second method is the CRC-NTMAC. It replaces the conventional MAC with a CRC (easy to compute, with predictable performance, but non-secure) and adds an encryption stage after concatenating the CRC's.
We've also extended the CRC-NTMAC by replacing the CRC with an error detecting code, obtaining the BCH-NTMAC.
Timestamping
Two graduate students and I are developing a distributed system for digital timestamping. In timestamping, a user submits a document (actually a cryptographic hash of the document) to a service. The service returns a timestamp and a cryptographic signature of the hash and timestamp. The user can then use the timestamp to prove that the document existed at that time.Our system avoids some of the problems of existing schemes--although, of course, introducing its own set of problems! We are currently testing an implementation.