Coverless steganography based on image retrieval of DenseNet features and DWT sequence mapping

Abstract

Recently, researches have shown that coverless image steganography can resist the existing steganalysis tools effectively. On this basis, a novel coverless image steganography algorithm based on image retrieval of DenseNet features and DWT sequence mapping is proposed in this paper. Firstly, DenseNet convolutional neural network model in deep learning is used to extract the features of image datasets. Supervised learning is adopted to retrieve the image, and the retrieval results can be used as the information carrier. Secondly, the selected images are divided into 4 × 4 sub-blocks for block discrete wavelet transform. Then the DWT coefficient are calculated based on the low-frequency components after block transformation, and the coefficients between blocks are scanned according to the Zigzag scan, such that the robust feature sequences are generated. Finally, the secret information is divided into segments with the same length as the feature sequence, and an inverted index is established with feature sequence, the position of blocks, DWT coefficient and image path. The image with the same feature sequence as the secret information segment is selected through index as the carriers. The experimental results and analysis show that this method has better robust and security performance resisting most image attacks compared with the state-of-the-art methods.