【作者】 马荣敏；

【导师】 孙斌；

【作者基本信息】 北京邮电大学 ， 计算机应用技术， 2011， 硕士

【摘要】 随着多媒体技术以及网络技术的迅猛发展,各种数码相机以及功能强大,简单易操作的图像处理软件的广泛应用,使得数字图像的修改越来越简单。在给人们日常生活带来方便的同时,也带来许多负面影响,如经过篡改的新闻图片或取证图像会给社会带来严重的不良影响以及妨碍司法公正。如何对数码图像的真伪进行有效地辨别,成为当今的研究热点。JPEG压缩格式以其特有的优点,成为当前图像的主流压缩标准,由于其在压缩比较高的情况下仍能保持较好的图像质量,因而被广泛应用于各领域,所以对JPEG图像真实性的研究具有重要的意义。本文在AC系数统计模型的基础上有针对性的提出并改进了三种基于JPEG二次压缩的数字图像真实性检测算法。在基于SVM的基础上设计并实现了AC系数拉普拉斯模型的JPEG图像真实性过滤算法,通过拟合图像一次JPEG压缩的概率曲线与图像实际的概率分布进行比较,提取特征向量进行SVM分类判别。使用非传统统计模型—均衡α稳态模型对AC系数进行描述,通过引入图像复杂度定义减少算法中特征指数值α的计算量,设计提取新的特征向量用于SVM分类判别。针对质量因子相近的二次JPEG压缩图像提出差分抖动检测模式,通过对AC系数直方图相邻系数做差分并进行抖动判定,来检测图像的真实性。实验证明,以上算法皆能有效地对非原始JPEG图像进行判别。更多还原

【Abstract】 With the rapid development of multi-medium technology and net technology, all kinds of digital cameras and powerful function、easy operation image process software are used widely. To tamper digital images is getting easier. While it brings convenience to our life, it also makes a lot of bad impacts. For example, tampered news pictures or modificated evidential images will result in serious bad affection and disturb the justness of judiciary. It’s becoming very important that how to distinguish the reality of JPEG images effectively.JPEG compress standard became the main image compress standard because of its special character. It can preserve better quality of images after a high rate of compress. So it is widely used in every fields. And it has an important signification to investigate the tampered JPEG images. This paper puts forward and improves three kinds of detection of double compressed JPEG images basis on the AC coefficients statistical model.Designing a filter algorithm for reality of JPEG image based on laplacian distribution. Obtain the eigenvectors by comparing the inosculation degree the simulative probability function of the one time compressed image’s AC coefficient and the probability function of AC coefficients of the image.Depicting AC coefficients with new statistical model—balanced a steadily model. Import the definition of image complexity to cut down the calculation of the value ofα. Design a new character vector for SVM to distinguish.Aiming at double compressed JPEG images which quality factors are very close, this paper designed difference twitter mode. To detect the reality of images by minus AC coefficients which are close to and determine whether they are twittered. Experiments show that all the algorithms are effective for detecting double compressed JPEG images.更多还原