Sparse Feature Adaptive Steganography in Color Image

In order to improve the performance of steganography in sparse domain,a steganography based on image components is proposed.First,two dictionaries are built,one for piecewise smooth parts(cartoon like) and the other for textures.Two methods of constructing dictionaries,using mathematical models and learned by K-SVD algorithm,are given.Then,the two dictionaries are combined to decompose the R,G and B channels,and sparse coefficients of the two kinds of image contents are obtained,respectively.Finally,secret information is embedded into nonzero coefficients of two channels,another channel is used to store decomposition path.During the embedding process,coefficients of textures are used to embed secret data in prior to those of piecewise smooth parts.Experimental results show that the proposed method can achieve high visual quality and outperform other existing sparse domain methods in anti-detection performance and robustness.