Accuracy increase for steganalysis methods by optimization of wavelet transform parameters

Abstract

Subject of Research. We have performed the study for accuracy increase of steganalysis methods used for detection of information hidden in wavelet domain. Methods. The methods proposed by Gireesh Kumar, Hany Farid, Changxin Liu, Yun Q. Shi demonstrate low efficiency when detecting the fact of embedding into the discrete wavelet transform domain. The paper considers the reasons for this fact. The basis of these methods is application of statistical moments for LL, HL, LH, HH domains obtained by wavelet transformation and additional image parameters making up a support vector. Main Results. We have analyzed the reasons for low efficiency of modern steganalysis methods in tasks of embedding fact detection in LH and HL domains that represents about 65% and 70% of correctly qualified images, respectively. The analyzed methods of steganalysis use such parameters as statistical moments, which have a high degree of variability that makes it difficult to classify images uniquely. We propose the technique for steganalysis accuracy increase by evaluation and taking account of the parameters’ error used by steganalysis methods by means of the specialized wavelet. The proposed technique gives the possibility to increase classification accuracy of images by 2.5–7.0 %. Practical Relevance. The results of work are useful for specialists in the field of information security in tasks of detection and countermeasures for the hidden transmission channels. The obtained results can be useful for development of steganalysis systems.