Extended Embedding Function for Model-Preserving Steganography

Abstract

Existing model-preserving steganography techniques are based on embedding data by modifying least significant bits of a cover image. To keep invariant component of pixel intact only, Least Significant Bit Replacement (LSBR) function is used to modify non-invariant component of the pixel. LSBR is found to be weak against visual or statistical attacks and provide limited steganographic capacity. Least significant Bit Matching (LSBM) embedding provides better security in comparison LSBR but it is not suitable for most of the model-preserving steganographic techniques. This paper explores the possibility of securely embedding data in least two significant bits of the cover image. It is shown that the embedding in least two significant bits violates the assumption of the structural steganalysis tools and techniques available to detect presence of a message in a stego image. Therefore, structural and non-structural detectors fail to detect presence of data in a stego image. The proposed embedding functions result in improved security and steganographic capacity in comparison to LSBM.