A secure adaptive Hidden Markov Model-based JPEG steganography method (J-HMMSteg)

This study introduces J-HMMSteg, an adaptive and secure JPEG image steganography technique designed for data embedding with minimal distortion. J-HMMSteg employed a block-wise analysis approach to detect shifts in image statistics and was performed in three phases. Firstly, it constructed statistical features of the images by analyzing intra-interblock correlations of quantized Discrete Cosine Transform (DCT) coefficients. Secondly, it utilized these features to create the Hidden Markov Model (HMM) of the images to capture complex characteristics such as smoothness, regularity, continuity, consistency, and periodicity. Finally, JHMMSteg utilized a maximum likelihood embedder driven by a threshold using Kullback-Leibler Divergence (KLD). The embedder maintained an optimal correlation by limiting the number of coefficients changed within a block according to its threshold; this ensured that only permissible distortions were introduced. This resulted in a stego image with minimal deviation from the HMM of the cover image's statistical distribution. The experimental analysis of J-HMMSteg was conducted on a database of 85000 JPEG images against four state-of-the-art approaches; the results demonstrated that J-HMMSteg was highly imperceptible, robust against RS steganalysis, and enhanced security against ensemble steganalysis.

Automated summary

This study presents J-HMMSteg, an adaptive and secure JPEG image steganography technique that embeds data with minimal distortion. J-HMMSteg utilizes block-wise analysis, statistical feature construction, Hidden Markov Model (HMM) creation, and a maximum likelihood embedder to achieve data embedding. Experimental results demonstrate that J-HMMSteg performs well in terms of imperceptibility, robustness against RS steganalysis, and enhanced security.