基于图像纹理复杂度的自适应隐写算法
首发时间:2018-03-26
摘要:结合人类视觉系统特性和图像纹理区统计特性,将密信自适应的嵌入纹理区可以获得更高的安全性,提出一种高效的自适应隐写算法,优选选择图像纹理区隐写,以G通道作为基准,根据纹理判别准则和隐写容量,自适应选择嵌入区域,其余两通道根据基准通道,可以获得对应的嵌入区域,在嵌入区域可以利用相邻像素的次LSB和LSB的关系实现密信隐写,提取阶段根据基准通道获得载密像素序列,将载密像素序列作用于二元函数提取密信。实验结果表明,以结构相似性和嵌入效率作为载密图像失真度的指标值均高于6种经典算法,提取通道相关性的R-SPAM、M-SPAM、A-SPAM特征并且利用集成分类器进行分类训练,预测的ROC曲线均优于LSBM系列算法,抗隐写分析能力更强。
关键词: 人类视觉系统 隐写容量 结构相似性 嵌入效率 隐写分析。
For information in English, please click here
An Adaptive Steganographic Algorithm Based on Image Texture Complexity
Abstract:Combining the characteristics of the human visual system with the statistical characteristics of the image textures, higher security of stego image can be obtained by adaptively embedding secret data in rich textures . As a result ,an efficient adaptive steganographic algorithm is proposed, and the textures of G channel is priority adaptively selected for embedding secret message according to the assessment criteria of texture complexity and steganographic capacity, and the remaining two channels can obtain corresponding embedding textures according to the G channel. In the embedding textures, the relationship between second LSBs and LSBs of adjacent pixels can be used to implement embedding of secret data.In extraction phase,according to the reference G channel to obtain the stego pixel sequence, the stego pixel sequence is applied to the binary function to extract secret information. The experimental results show that the structural similarity index and embedding efficiency that is used to measure image distortion are higher than the six classical algorithms , and the R-SPAM, M-SPAM and A-SPAM features of correlation channel are extracted from three channels , which is used to train and classify according to the ensemble classifier ,the predicted ROC curves are better than the based LSBM algorithms, and have stronger resistance to steganalysis.
Keywords: texture complexity steganographic capacity structural similarity index embedding efficiency steganalysis.
引用
No.****
动态公开评议
共计0人参与
勘误表
基于图像纹理复杂度的自适应隐写算法
评论
全部评论0/1000