Koshkina N. Spectral methods of computer steganography and methods of steganalysis with learn-ing and classification

The dissertation is devoted to solving the information security problem with using of methods of computer steganography and steganalysis. New robust spectral steganographic methods for marking digital audio signals and images are proposed and existing methods are improved. Their imperceptibility and robustness to common processing operations were substantiated and analyzed. The problems of watermark desynchronization and information protection on paper carriers were examined. Universal (blind) statistical steganalytical methods with learning and classification were improved and adapted to the new types of stego objects and tested under the different conditions. These methods are able to detect stego objects that have been created by publicly available steganography programs. Steganographic and steganalytical software for these methods was developed and numerical estimations of the quality main characteristics of a method were obtained. Key words: information security, computer steganography, spectral methods, discrete Fourier transform, wavelet transform, steganalysis, statistics, support vector machine.