For example,Бобцов

EFFECTIVENESS OF STEGANALYSIS BASED ON MACHINE LEARNING METHODS

Annotation

Subject of Study.The paper presents comparative accuracy estimation of modern machine learning-based steganalytic methods. The paper deals with the most perspective methods in tasks of the passive counteraction to the information transfer channels using the discrete wavelet domain of static digital images. Methods. We have studied methods proposed by Gireesh Kumar, Hany Farid, Changxin Liu, Yun Q. Shi and the SPAM method. Basically the methods apply statistical moments obtained from wavelet bands LL, HL, LH and HH, as well as additional image features forming a support vector. BOWS2 image collection was used to estimate the effectiveness of methods. Steganographic impact was modeled by changing the least significant bits of coefficients for the each DWT band (LL, LH, HL and HH) with 5% and 20% of payload. The effectiveness of explored methods is estimated in view of obtained true positive, true negative, false positive and false negative image classification values. Main Results. The study has shown that all explored methods except for SPAM are effective in the task of detecting of embedding in HH band. As for the detection of the embedding in LH band, Yun Q. Shi is the most effective algorithm. In the task of the detecting in HL band, all explored methods except for SPAM have appeared to be comparatively effective under condition of big payload. When detecting the embedding in LL band, all methods have shown the effectiveness about 50% regardless the payload rate. It is established that the considered methods are not able to render effective counteraction to the hidden data channel, using the LH and HL region due to the fact that they use Haar wavelet transform. It is concluded that the application of the optimal wavelet transform makes it possible to reduce the intersection area of value histograms of the first statistical moment for the original images and steganoimages. Practical Relevance. The work results are useful to the specialists in the field of information security in the tasks of detection and combating the hidden data channels. The obtained results can be used in the development of steganalysis systems and improved methods of steganalysis as well.

Keywords

Постоянный URL

Articles in current issue

CONTENT © ITMO UNIVERSITY 1993-2024