Development of digital communications systems significantly extended possibility to perform covert communications (steganography). This recalls an emerging demand in highly efficient counter-measures, i.e. Steganalysis methods. Modern steganography is presented by a broad spectrum of various data-hiding techniques. Therefore development of corresponding Steganalysis methods is rather a complex problem and challenging task. Moreover, in many practical Steganalysis tasks second Kerckhoff's principle is not applicable because of absence of information about the used steganography method. This motivates to use blind Steganalysis, which can be applied to the certain techniques where one can specify at least statistics of the hidden data. This paper focuses on the class of supervised Steganalysis techniques developed for the additive steganography, which can be described as y = f(x, s, K) = x + g(s, K), where stego image y is obtained from the cover image x by adding a low-amplitude cover image independent (±1 embedding also known as LSB matching) or cover image dependent (LSB embedding) stego signals that may be also depended on secret stego key K and the secret data s. The function g(.) represents the embedding rule. The proposed method provides the stochastic interpretation of the blind Steganalysis and consists of two main stages, i.e., data preprocessing and feature extraction. The data preprocessing targets at stego signal estimation that is performed in the wavelet domain from the mixture of cover image (presented by non-stationary Gaussian model) and stego signal (presented by stationary Gaussian model). Feature extraction is realized using model-based (polynomial) approximation of stego image pdf. In this case polynomial coefficients, which simultaneously are high order statistics, have created the feature set. Because the features are calculated from the estimated stego signal, they are more sensitive to steganographic modifications while suppressing the influence of the cover image.
展开▼
机译：数字通信系统的开发显着延长了执行隐蔽通信（隐写术）的可能性。这回顾了高效的反措施的新兴需求，即塞巴巴分析方法。现代隐写术由广泛的各种数据隐藏技术提出。因此，相应的隐分方法的发展是一个复杂的问题和具有挑战性的任务。此外，在许多实际的隐星分析任务中，第二克雷克霍夫的原理是不适用的，因为没有关于使用的隐写法方法的信息。这激励用于使用盲人瘫痪，这可以应用于某些技术，其中一个技术可以指定至少统计隐藏数据。本文重点介绍了为添加剂隐写术开发的监督隐分技术，其可以描述为y = f（x，s，k）= x + g（s，k），其中stego图像y是从盖子获得的通过添加低幅度覆盖图像的图像X独立（±1嵌入也称为LSB匹配）或覆盖图像相关（LSB嵌入）STEGO信号，其也可以依赖于秘密的SEGO密钥K和秘密数据。函数g（。）表示嵌入规则。该方法提供了对盲体分析的随机解释，包括两个主要阶段，即数据预处理和特征提取。在从覆盖图像的混合（由非静止高斯模型）和Setego信号（由静止高斯模型呈现）中，在小波域中执行的数据预处理目标。使用基于模型（多项式）近似的STEGO图像PDF实现特征提取。在这种情况下，同时是高阶统计的多项式系数已创建该特征集。因为这些特征是从估计的SEGO信号计算的，所以它们对隐点修改更敏感，同时抑制覆盖图像的影响。
展开▼
