This text will develop and formalize the applications of machine learning in steganalysis. Researchers in steganalysis typically have a background in signal and image processing, or multimedia coding, and lack the high level statistical knowledge required for using machine learning; this book will provide an accessible introduction to the subject, as applied specifically to steganalysis. Much of the understanding of machine learning that can be gained from this book can be adapted for future study of machine learning in other applications.
The book begins with an overview of current systems and theory within steganalysis, followed by an introduction to the concepts and uses of machine learning. Feature selection and classifiers are then discussed in depth. The coverage of feature selection will include evaluation, completeness, wavelet features, JPEG features, and a brief exploration of less commonly known feature sets. The section on classifiers will cover Support Vector Machines, kernel methods in general, non-learning classifiers, ANN, unsupervised learning, and evaluation heuristics. Whilst the focus is on steganalysis, it will include some steganographic methods (eg F5, YASS) where these are necessary to evaluate steganalytic techniques. Whilst steganalysis and machine learning are integrated throughout the book and illustrated with numerous examples, a final section brings them together to examine training set selection, fusion in steganalysis, and comparisons and evaluations. The book concludes with discussion of future directions of the discipline, and the role of machine learning in steganalysis as part of the wider trend on digital forensics.
Theory will be discussed in the text, supported by complete source codes provided in an appendix and on a supporting website.
Keywords: Signal Processing