稀疏特征自适应的彩色图像隐写

Sparse Feature Adaptive Steganography in Color Image

为了提高稀疏域隐写的性能,提出一种基于图像成分的稀疏域隐写算法.首先构造2个字典,分别用于稀疏表示图像的分片平滑成分(卡通成分)和纹理成分,并给出了2种构造字典的方法,一种是利用现有数学模型,另一种是利用K-SVD算法进行自适应学习;然后结合2个字典对彩色图像的R,G,B通道进行稀疏分解,分别获得2种图像成分的稀疏表示系数;最后将秘密信息嵌入到其中2个通道的非零表示系数中,并优先选择纹理成分稀疏表示系数,另一通道则用于保存分解路径.实验结果证明,该算法在获得较高视觉质量的同时,比其他稀疏域隐写算法具有更强的抗隐写分析能力和更好的鲁棒性.

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.