Watermarking, steganography and content forensics

Electronic watermarking is about 60 years old. However it was not until the beginning of the early 1990's that watermarking received widespread interest, due to concerns about piracy of digital content. In the subsequent decade, very significant progress has been made both in our theoretical understanding of digital watermarking and in its applications. In this talk we highlight some of the progress made and describe a number of current and possible watermarking applications. Steganography has a much longer history, dating back to at least the time of the ancient Greeks. While Shannon dismissed steganography as "primarily a psychological problem", the last decade has seen the application of information theory to steganography. Terrorist events at the beginning of the 21st century motivated further attention and very interesting results have been described. At first sight, digital watermarking and steganography would appear to share the same goals. However, while both seek to hide information within other information or content, there are very significant differences in the constraints that must be satisfied. For example, in digital watermarking being imperceptible is very important. However, in steganography being imperceptible is less important than being undetectable, since the original cover Work is not available for comparison. The similarities and differences between digital watermarking and steganography are highlighted here. Steganography spawns steganalysis, the art and science of detecting the presence of a steganographic message hidden in innocuous content. Recent research views steganalysis as a binary classification problem; is a hidden message present or absent? Classification is based on testing for statistical anomalies in features derived from the content. We will describe a variety of steganalysis algorithms, and discuss approaches to evaluating and benchmarking these algorithms. Content forensics shares similarities with steganalysis. At the simplest level, content forensics is often asked whether, for example, an image is authentic or has been tampered with. This problem can also be viewed as a binary classification problem and similar techniques can be applied. We will review recent work in content forensics and discuss their current limitations.