The ISC International Journal of Information Security

Abstract This work proposes a novel steganographic scheme that employs deep learning to embed RGB images into audio files, and it also introduces an innovative steganalysis approach. The proposed method embeds an image at the bit level within the audio in the frequency domain, enhancing flexibility for embedding various data types. The network uses an encoder-decoder architecture, where the encoder embeds bits into the audio, and the decoder extracts the embedded bits from the audio. To enhance the information transmission rate, an image compression method based on the YUV color model is used. This method can reduce the data to be hidden and transmitted for the image by up to 50%. The steganographic encoder-decoder architecture incorporates multiple paths to facilitate gradient flow and network training. The proposed steganalysis network effectively detects stego audio files containing hidden messages by analyzing the signal’s transform domain features. The results demonstrate the proposed steganography scheme’s enhanced security, with audio Signal-to-Noise Ratio (SNR) ranging from 26.9 to 39.6 dB and image Peak Signal-to-Noise Ratio (PSNR) from 19.02 to 34.8 dB. Compared to other audio steganography schemes, the proposed method is shown to have a higher performance in terms of audio cover perceptibility and hidden image quality.

Abstract The aim of image steganalysis is to detect the presence of hidden messages in stego images. We propose a blind image steganalysis method in Contourlet domain and then show that the embedding process changes statistics of Contourlet coefficients. The suspicious image is transformed into Contourlet space, and then the statistics of Contourlet subbands coefficients are extracted as features. We use absolute Zernike moments and characteristic function moments of Contourlet subbands coefficients of the image to distinguish between the stego and non-stego images. Absolute Zernike moments are used to examine the randomness in the test image and characteristic function moments of Contourlet coefficients is used to form our feature set that can catch the changes made to the histogram of Contourlet coefficients. These features are fed to a nonlinear SVM classifier with an RBF kernel to distinguish between cover and stego images. We show that the embedding process distorts statistics of Contourlet coefficients, leading to detection of stego images. Experimental results confirm that the proposed features are highly sensitive to the change made by the embedding process. These results also reveal advantage of the proposed method over its counterpart steganalyzers, in cases of five popular JPEG steganography techniques.

Abstract So far, various components of image characteristics have been used for steganalysis, including the histogram characteristic function, adjacent colors distribution, and sample pair analysis. However, some certain steganography methods have been proposed that can thwart some analysis approaches through managing the embedding patterns. In this regard, the present paper is intended to introduce a new analytical method for detecting stego images, which is robust against some of the embedding patterns designed specifically to foil steganalysis attempts. The proposed approach is based on the analysis of the eigenvalues of the cover correlation matrix used for the purpose of the study. Image cloud partitioning, vertical correlation function computation, constellation of the correlated data, and eigenvalues examination are the major challenging stages of this analysis method. The proposed method uses the LSB plane of images in spatial domain, extendable to transform domain, to detect low embedding rates-a major concern in the area of the LSB steganography. The simulation results based on deviation detection and rate estimation methods indicated that the proposed approach outperforms some well-known LSB steganalysis methods, specifically at low embedding rates.