图像自适应隐写的可靠检测

Adaptive image steganography is a technology that hides the secret messages into the hard-to-detected area based on the content of the images. Because it performs better than the typical steganographic method, it attracts many information hiding researchers in recent years. Generally, it is hard to model the hiding area of adaptive steganography, and the noise caused by adaptive steganography is similar to the noise caused by some common image processing (for example, sharpening operation). Thus, how to increase the reliability of adaptive steganalysis is a valuable and important question. To solve the question, this project will researches from four topics: feature extraction, classification of PhotoShop (PS) images, feature selection and secret information extraction. First, by evaluating the ability of various filters on capturing the modifications caused by embedding and the rationality of weighting for feature extraction samples, the feature extraction method based on filters selection and feature extraction sample weighting for adaptive steganalysis will be researched. Second, by analyzing the statistic influence made by image processing, the steganalysis method based on operation characteristic will be proposed. Third, by quantifying the dependency between feature components and attribute significance of feature components, the selection method based on the measurement of separability between feature components will be researched for steganalytic features. Forth, the characteristics of the parity check matrix foundation and stream encryption in the Syndrome Trellis Codes (STCs) are utilized in the exploration of extraction method for adaptive steganography. It is expected to increase the reliability of adaptive steganalysis based on the research topics in this project, and the results will promote the development of information hiding area.

图像自适应隐写主要在纹理复杂区域或者更改后失真较小的位置进行更改，已经成为信息隐藏方向的研究热点之一。图像自适应隐写更改的区域或位置建模通常较为困难，而且自适应隐写噪声与部分正常处理操作（如锐化）的噪声非常相似。这使得现有检测方法的可靠性有待于提高。本课题拟从特征提取、PS图像分类、特征选取和信息提取等方面出发，重点开展下列研究：考虑不同滤波器对隐写更改的捕获能力和特征统计样本权重分配的合理性，研究基于滤波器选取与特征统计样本权重分配的自适应隐写检测特征提取；考虑常用图像处理操作对图像统计特性的影响，研究基于处理操作特性的自适应隐写检测；考虑特征分量间的相关性和特征分量属性的重要度，研究基于特征分量可分性度量的隐写检测特征选取；利用校验矩阵的构成特点和流加密特性，探索基于STC编码特点的自适应隐写信息提取。上述研究有望进一步提高图像自适应隐写检测的可靠性，促进信息隐藏技术的发展。

图像自适应隐写主要在纹理复杂区域或者更改后失真较小的位置进行更改，已经成为信息隐藏方向的研究热点之一。由于对纹理复杂区域或者更改后失真较小位置建模较为困难，且常用的对比度增强、锐化等操作在图像中添加的噪声与自适应隐写噪声具有非常相似的统计特性，对图像自适应隐写进行可靠检测十分困难。本课题围绕图像自适应隐写的可靠检测，从下列多个方面开展研究，并取得了系列成果。在基于特征改进的图像自适应隐写检测方面，提出了基于离散傅里叶变换的特征去冗余方法、基于图像纹理模式编码的隐写检测特征提取方法、基于通道梯度相关性的彩色图像隐写检测方法和基于差分通道嵌入更改概率的彩色图像隐写检测特征提取方法；在基于PS操作特性的图像自适应隐写检测方面，提出了基于直方图零点位置的伽马变换参数估计方法、基于伽马变换性质的图像操作检测方法、基于阈值化二元编码的图像锐化检测方法、基于图像操作识别的隐写检测方法、结合图像操作鲁棒检测的隐写检测框架；图像隐写检测特征选取方面，基于改进Fisher准则的特征选取方法、基于全局特征MMD差值的Rich Model隐写检测特征选取方法、基于特征分量相关性的Rich Model隐写检测特征选取方法、基于局部特征MMD增益的Rich Model隐写检测特征选取方法、基于大规模图像数据预分类和特征选取的自适应隐写检测方法；图像隐写信息提取方面，基于最优隐密像素子集的MLSB隐写信息提取方法、基于提取信息比特分布差异的隐写密钥恢复方法和基于信息分布差异的F5隐写密钥还原算法。此外，还在3D图像隐写检测、图像鲁棒隐写设计等方面开展了相关研究。上述研究结果提高了对隐写检测的可靠性，为促进隐写取证的实用化提供了技术上的支持。.已形成并在“Signal Processing”、“Signal Processing: Image Communication”和“计算机学报”等发表相关论文34篇，其中SCI期刊26篇；申请国家发明专利11项，获授权3项。培养博士生3名，硕士生4名。研发的图像隐藏信息智能监测系统已在军队和国家安全单位的网络安全防护和测试中投入试用。