Constructive image steganography using example-based weighted color transfer

Abstract

Image steganography aims to hide secret information without arousing suspicion that an image has conveyed hidden messages. The majority of image steganographic algorithms usually embed secret messages by altering pixels of a cover image. Despite achieving some degree of success, such a cover modification approach leads to certain telltale signs detected by steganalytic tools, thus unveiling the very existence of secret messages. In this paper, we propose a novel constructive steganographic algorithm to disguise the hidden information without involving a pre-existing cover image. Our WCTCIS scheme transforms secret messages into a stego image, taking advantage of example-based weighted color transfer, which synthesizes a new image with different colors by referring to a reference image. Our scheme first encodes secret message to increase the security using a modified two-dimensional lag-complex logistic map. An optimal weight is then derived using one of four measure techniques we suggest. Next, we conduct an example-based weighted color transfer to produce a temporary image, where each pixel is represented by a floating-point format. Finally, we conceal secret messages when synthesizing a pixel using a predetermined embedding vector. An intensive testing based on a database containing 5,000 natural images confirms that our algorithm offers a large payload and produces a stego image with significantly high quality. The CIEDE2000 color difference metric demonstrates that the stego image has high-fidelity color appearance. A security analysis shows that our scheme can resist current state-of-the-art learning-based ECMV steganalysis as well as statistic-based RS steganalysis, and is robust to salt-and-pepper noise attacks. The probability of breaking down our scheme is insignificantly small. The contribution of our work is introducing a constructive steganographic algorithm which makes use of color changes to conceal secret messages, produces better performance than the current state-of-the-art algorithms, and extends a color processing technique to achieve constructive steganography.