隐写是通过利用将秘密消息隐蔽在可公开的内容中传递给特定的接收者而不为第三方所察觉的技术。随着互联网的普及,隐写安全及与其相对的隐写分析成为信息安全领域研究的热点之一。图像是隐写中最广泛使用的载体媒介,图像纹理对隐写隐蔽性具有重要影响。本文以图像纹理特征对隐写及其安全测评分析的影响和利用为研究内容，针对目前在图像纹理差异对隐写安全测评的影响以及如何利用图像纹理特征改进隐写安全性上研究的不足，筛选对隐写敏感的纹理特征，设计有效的图像纹理分类方法并分析图像纹理差异对隐写安全性影响。最后，基于纹理特征相似性研究隐写载体选择方法改进隐写安全性。

本文在详细调研现有各种图像纹理特征的基础上，分析其对不同隐写算法敏感程度，筛选和发现对隐写有效的图像纹理特征，仿真实现了5种纹理特征提取算法，并全部集成到“面向隐写的图像纹理分类系统”实现中。然后，提出了一种新的针对隐写的图像纹理复杂度度量方法用于衡量图像间纹理差异，并设计实验基于纹理复杂度对图像进行分类，详细分析图像纹理差异对隐写分析性能的影响。结果表明客观地评价隐写安全性应测试其在不同纹理特征图像测试集上的性能表现。其次，训练集的构造同样对检测准确率有着重要的影响，训练图像集越“完备”，隐写检测率越高。在分析某一纹理复杂度图像时，使用纹理复杂度相近的训练集，可以得到更高检测准确率。最后，本文分析了载体选择问题应用的三种场景，提出了一种通用的隐写框架，将用于嵌入过程前后的载体选择、嵌入过程中的失真最小化控制以及嵌入操作完成后的隐写特征还原阶段集成其中。对比实验结果表明，应用载体选择方法在隐写过程中可以大幅提高隐写安全性。不过，其中对衡量隐写分析特征失真度量准则的预先选择在很大程度上影响了载体选择的有效性。

Steganography is the art and science of writing hidden message within the public media that no one apart from the sender and intended receipt suspects the existence of the message. Along with the wide spread of Internet, steganography and steganalysis have become a hotspot in the research field of Information Security. Image is the most widely used cover media in information hiding, and image texture have an important impact on steganographic security. Due to the lack of pointed and systematic analysis on image texture in the this field, this paper will mainly focus on the impact of image texture feature to steganography, Sensitive texture features will be chosen and used to analyze the influence of image texture on steganographic security and steganalysis performance. At last, a new cover selection method based on similarity of image texture feature is proposed to improve the steganographic security.

Firstly, we will test the sensitivity of various image texture features to different steganographic algorithms based on preliminary research results, choose and discover useful ones among them. Five different image texture extracting algorithms are implemented and integrated to the “Image Texture Classification System to Steganographic Application”. Then, the impact of image texture diversity on steganography is detailedly analyzed in this paper, we present a new metric of image texture complexity based on the high dimensional image statistical model to measure the texture difference of images. Extensive experiments are conducted to make comparison of the detection performance under different embedding rates and image complexities. The comparative experimental results indicate that the texture difference of image is another critical factor for evaluating detection performance. In addition, the composition of the training set play a vital role in the generic steganalysis process, the more completed training set will get the higher detection accuracy. On detecting images with some specified texture complexity, better detection rate of accuracy will be obtained while training set having close complexity is used. At last, three possible scenarios in which cover selection problem applies are investigated. A generalized steganographic framework integrated with cover selection (CS), Restoration Minimization (DM) and Feature Restoration (FR) is proposed in the paper. Experiment results show that cover selection method introduced in this paper greatly minimizes the chances of detectability by the steganalyzer. However, the chosen metric for the distortion of feature vector is crucial to the effectiveness of cover selection method.