Multi-layer assignment steganography using graph-theoretic approach

Abstract

This paper proposes a novel multibit assignment steganographic scheme for palette images, in which some colors in the palette are exploited to represent several secret bits. For the proposed scheme, each palette color is treated as a graph vertex, and an edge among any two vertices indicates an adjacent relationship between them. A graph traversal technique named depth-first search is used to accomplish the multibit assignment for the vertices that correspond to the palette colors. The major idea of the proposed data-embedding is to modify colors of image pixels according to the assigned bits and secret message. Image pixels are classified as embeddable pixels and non-embeddable pixels before data-embedding. During the data-embedding, for each embeddable pixel, if the original color of the pixel matches the secret data, the pixel is then kept unchanged; otherwise, a suitable adjacent color of the original color will be used to replace the original color so that the new color matches the secret data. Experimental results show that the secret data can be embedded and extracted successfully without introducing visual artifacts, and the proposed scheme can always achieve a high capacity or maintain good image quality, comparing with the related works.