Color image steganography scheme based on FLD ensemble classifiers
Mingming Jiang, Guangming Tang, Yi Jie Sun, Shunxiang Yang · 2017
In order to preserve the statistical properties of information hiding when embedding changes are constrained to pixels, a color image steganography method is proposed based on FLD Ensemble classifiers and content-adaptive quadtree algorithm. First, the color cover is divided into several non-overlapping sub-blocks by quad-tree algorithm, and each block is prioritized depending on its texture complexity. Then, we extract the statistical feature of blocks with high priority from three channels respectively. Then, we use such feature as the input of FLD Ensemble classifiers, which are trained by color stego images generated by the current steganography schemes LSBM, WOW and S-UNIWARD. Then, we design the additive distortion function of color images and the pixel distortion is computed by results of the Ensemble classifiers. We use STC embed secret message into R, B, G channels respectively. The experimental results show that the proposed scheme can achieve a competitive performance compared with the other steganography schemes LSBM, WOW and S-UNIWARD when resisting SPAM steganalysis methods for color images effectively.