Key Technology Research On Coverless Steganography Based On Image Selection

Steganography is a vital technology to protect secret information.The traditional information embedding-based image steganography usually embeds secret information into the image without visual changes to achieve the purpose of secret information protection and covert communication.However,some modification traces will be left in the image after embedding secret information,which may be detected successfully by the steganalysis techniques,thus causes certain security risks.To fundamentally resist the detection of steganalysis technology,image selection-based coverless steganography has attracted increasing attention,which provides security guarantee for covert communication and is significant for the secret information protection and the development of the information security field.However,the current image selection-based coverless steganography methods have a wide range of problems,such as low secret message hiding rate,low effective capacity,more difficult to construct the coverless image dataset,etc.,which limits the further development of this field.To solve these scientific problems in the field of image selection-based coverless steganography,some key technologies such as inconsistent mapping,constructing coverless image dataset by unsupervised clustering,dual-constructing small coverless image datasets,multiple features combination-based optimization strategy,etc.are proposed in this paper.The following methods have been proposed:(1)Coverless Steganography Based on Inconsistent Mapping of Sub-image’s Average PixelTo improve the secret message hiding rate and the effective capacity,coverless steganography based on inconsistent mapping of sub-image’s average pixel is proposed in this paper.In the proposed method,a hash dictionary is built firstly to store the sub-secret message sequences.In this hash dictionary,each sub-secret message is mapped to one label.Then,the cover image is divided into several sub-images,and the average pixel values of sub-images are used to generate a hash sequence to represent the image.The dimensional of the generated hash sequence is larger than that of the secret message.Hence,to hide the low-dimensional secret message by the high-dimensional hash sequence,an inconsistent mapping rule is designed finally to relate the high-dimensional hash sequence to the low-dimensional secret message.Experimental results show that the given image dataset can be mapped to more secret message sequences through the inconsistent mapping rule,which improves the secret message hiding rate.Besides,a relatively large-size coverless image dataset can be constructed,which can achieve a higher effective capacity.In addition,the relationship between the construction of a coverless image dataset with a specific size and the number of images required in the proposed method is analyzed,which lays a foundation for subsequent research.(2)Coverless Steganography Based on Unsupervised Clustering and Inconsistent MappingTo decrease construction difficulty of coverless image dataset and improve robustness of coverless steganography,coverless steganography based on unsupervised clustering and inconsistent mapping is proposed in this paper.In the proposed method,high-dimensional deep hashes of the given image dataset are extracted firstly.Compared with low-dimensional secret messages,high-dimensional deep hashes have more expressions,which lays the foundation for constructing a large-size coverless image dataset.Then,the unsupervised clustering algorithm is adopted to construct the coverless image dataset in this paper,which makes it simple and efficient to construct the coverless image dataset.The deep hashes between the images in the coverless image dataset constructed by the unsupervised clustering algorithm have relatively large distances,which can resist strongly the image processing algorithms.What’s more,the construction of the coverless image dataset in this paper can be duplicated on both the sending and receiving sides,which avoids the potential risks brought by transmitting the coverless image dataset.Finally,an inconsistent mapping rule is built to relate the high-dimensional deep hashes of coverless image dataset to the low-dimensional secret message sequences,which can map any different images into different secret message sequences.Experimental results show that the proposed method not only achieves higher robustness,but also can construct a large-size coverless image dataset more easily,and thus achieves higher effective capacity.(3)Coverless Steganography Based on Message Division and Image SynthesisTo reduce the size of coverless image dataset that needs to be constructed when hiding the same number of secret message sequences and further improve the hiding capability of coverless steganography,coverless steganography based on message division and image synthesis is proposed in this paper.In the proposed method,one secret message is firstly divided into two sub-secret messages.Then,two small coverless image datasets are constructed correspondingly and the inconsistent mapping rules are built to related the two small coverless image datasets to the two sub-secret messages.The construction of two small coverless image datasets can be also duplicated on both the sending and receiving sides,which avoids the potential risks brought by transmitting the small coverless image datasets.Finally,two stego-images are synthesized into one camouflage image to decrease the number of images that need to be transmitted.In addition,multiple robust features are combined and an optimization method is designed to improve the robustness.Experimental results show that the proposed method breaks the exponential relationship between the size of the coverless image dataset and the length of secret message.Therefore,the smaller size of the coverless image dataset can be constructed to hide the same number of secret message sequences,which further improves the hiding capability of image selection-based coverless steganography.