Watermarking for Identification and Change Detection of Sensor Data

Sensor data containing sound, images and video with various frequency contents are important, and will in the future be even more so, within the Swedish Defence. Watermarking is a method which can be used to embed information in such data. After the embedding, there should be no conceptual difference between the original and watermarked sensor data. The watermark can at a later stage be extracted from the sensor data and used for security applications. In this report we will we look into two such applications: 1. Identification - A watermark is embedded in the sensor data in a robust manner. It can contain important information like, for example sensor id., timestamps, sensor settings, GPS data and copyright information. The reason for this information to be stored as a watermark rather than in a header is that it is harder to manipulate the information afterwards. In addition, the information will remain even after a format change or another transform, which means that important data is not lost. 2. Change detection - To digital sensor data there is no ´negative´ which can be used to prove what the original data looked like. Therefore, there is a risk that deliberate manipulations or accidental changes of sensor data results in decisions made on false premises. By using so called fragile watermarks, it is possible to decide whether sensor data have been changed or not. With some fragile watermarking methods, it is in addition possible to discover in which parts for example an image have been manipulated, and some methods can even partly recover the original information in these areas. This report starts by addressing the fundamentals of watermarking, and is then studying achievements within the above mentioned application areas. A scenario with surveillance cameras around a restricted area is used to illustrate how watermarking can be used in different ways to improve the security.