Advanced steganographic and steganalytic methods in the spatial domain

In the first part of this dissertation, we introduce a detection and message length estimation method targeted at PM1 steganography used in spatial domain of digital images. We model the act of embedding as noise adding and develop a maximum likelihood estimator of the unknown message length. The performance of the method depends greatly on the source of images used for embedding. It works accurately for images with low amount of noise such as JPEG compressed images and images taken by digital cameras---images most frequently used in personal communication. In the second part of this work, we discuss a special kind of adaptive embedding---the Perturbed Quantization (PQ) steganography---as a means of improving the security of PM1 embedding. PQ is an embedding technique based on the premise that by using side information, not available to the recipient (and thus an attacker), we should be able to achieve much better security. The implementation of PQ uses a concept of wet paper codes---codes that allow communication without the need for sharing the selection channel (the placement of the hidden message). The wet paper codes are interesting in their own right because they solve many of the problems that any adaptive embedding faces. In the final part of this dissertation, we describe a method for minimization of the number of embedding changes. We argue that the number of changes an embedding technique does during embedding greatly influences the detectability of the technique. This leads to the concept of embedding efficiency, which is defined as the number of random message bits embedded per one changed pixel. It has been shown that there is a connection between the concept of embedding efficiency and the concept of covering codes from coding theory. We use this connection to derive an upper bound on the embedding efficiency. (Abstract shortened by UMI.)