【作者】 赵俊红；

【导师】 朱学峰；

【作者基本信息】 华南理工大学 ， 模式识别与智能系统， 2011， 博士

【摘要】 随着计算机技术的发展,作为其分支之一的数字图像处理技术的发展也是日新月异,新的数字图像处理技术和相应的软件不断地涌现,其功能变得越来越强大而操作同时变得越来越“傻瓜”。越来越多的非专业摄影人士也能熟练使用图像处理软件。以前需要在暗房中经过繁琐的操作才能完成一幅图像的篡改操作,而现在只需要一台普通的计算机和功能强大的数字图像处理软件就可以让一个普通的非专业人士完成一幅图像的篡改。一个熟练的操作者能让图像不留下肉眼能察觉的痕迹,达到以假乱真的目的。对图像真伪的判断光靠肉眼是看不出任何端倪的。因此图像处理产生了一种新的技术——数字图像被动取证技术(也称盲取证技术)。数字图像被动取证技术就是利用各种检测手段对图像进行检测和研究,从而判断出图像是否被篡改、是否来自同一获取设备或者是否计算机生成等。图像被动取证技术与图像主动取证技术最大的区别在于:图像主动取证技术需要事先往图像中添加数字水印或者数字签名信息,然后通过检测数字水印或数字签名的完整性来判断图像是否受到了篡改;而被动取证不需要事先添加任何信息到图像中,而是直接利用图像自身信息来进行鉴别和取证,因而被动取证具有更广阔的应用前景。由于图像被动取证涉及范围很广,如针对图像真伪,针对图像来源,针对图像是否计算机生成等应用领域,不可能一应俱全都研究到。本文针对图像内容被动取证进行了较深入的研究,主要研究内容和取得的研究成果如下:1.针对现阶段图像被动取证技术没有一个公认的确定的理论框架结构这个问题,在查阅大量的参考文献基础上,结合开闭环控制思想,初步构建了一种图像被动取证的框架;在该框架中首先利用标准图像数据库和质量评价指标不断调整优化特征提取方法,使取证算法所使用的特征量能更准确的刻画图像篡改部分的本质,同时也对图像取证算法进行不断优化,得到优化后的特征提取和优化后的取证算法;再利用优化后的特征量和优化了的取证算法对未知篡改类型的图像进行被动取证;2.针对现阶段的被动取证技术手段都比较单一,尚没有一个确切的分类方法,本文提出了一种被动取证技术的分类方法,将其分为:基于图像内容的被动取证、基于成像过程的被动取证和基于物理原理的被动取证;3.图像篡改中为了掩饰篡改痕迹往往会对图像中物体的边缘进行模糊操作,本文对此展开研究,将模糊数学和粗糙集理论引入其中,提出了一种基于模糊增强和粗糙集的新算法,结果表明该算法能较好的检测出被人工模糊操作的边缘;4.将水平投影和垂直投影引入到图像被动取证中,提出了基于水平投影和垂直投影的图像被动取证方法。与Posucue算法相比表明了,新算法具有更快的速度;还引进了改进后的圆投影,试验结果证明,利用改进的圆投影可以克服水平垂直投影不能抗旋转的缺点,对角度旋转,光照变化都具有一定的鲁棒性;5.首次将仿射尺度不变特征变换算法(ASIFT)用于图形复制--粘贴型篡改中,提出了ASIFT的图像复制—粘贴型篡改取证方法,试验结果表明,该算法在图像深度旋转,大尺度变化时有较好的表现;首次将流形算法中的局部线性嵌入算法(LLE)用于复制—粘贴型篡改检测中,结果表明,跟用主元分析法降维方法相比,基于LLE的算法对图像的模糊操作具有一定的鲁棒性,同时在原有的LLE算法上进行了改进,使得LLE处理速度得到一定程度的提高。最后在总结本文研究的基础上,对图像被动取证的未来发展方向进行了展望。更多还原

【Abstract】 With the development of the computer technology, digital image processing technology, as one branch of computer science, is developing fast. New technology and new software of image processing have come forth with more convenient manipulation. More and more people can modify the pictures only with a computer and image processing software such as Photoshop and 3DMAX. A modified picture can not be recognized by eyes because of lacking of the tampering traces. Passive image forensics has been developed which detects the image by many methods to find out the fact whether the picture is modified or not. Meanwhile, it can also tell us that the picture is natural image or a computer generate.The most difference between passive image forensics and active one is that digital signature and digital watermark should be embedded in images in advanced in active forensics. The passive forensics uses the information which is applied by the indeed images. Then, the passive one will have more applications.This paper reveals the research on passive image forensics based on image content. The main achievements reveal as following:Aiming at the question that there is no acknowledged theoretical framework of passive forensics, it gives a preliminary framework.Aiming at the problems that every method revealed is detecting one certain tamper, there is no one certain classification method. Then it classified all methods into three classifications which are passive forensics based on image content, passive forensics based on the capturing process and passive forensics based on physical principles.Aiming at edge flurring, the general manipulation in image tampering, a new algorithm has been applied which uses fuzzy enhancement and rough sets to find out the edges which is flurred manually. The result shows the algorithm works well.Applying projection data to passive image forensics, a new algorithm is more rapid than Posucue’s algorithm. Furthermore, it improves the algorithm by replacing horizontal/vertical projection to Ring projection which makes the algorithm rotation invariant and robust to illumination.ASIFT algorithm has been applied in detection the copy-paste forgery at the first time. The result shows that the algorithm can work well when the image rotates and large scale changes. LLE algorithm has been applied in detection the copy-paste forgery at the first time. LLE is some robust to edge flurred. An improved LLE shows that the improved one is more rapid than original LLE.Finally, the paper summarized the research works and gave the prospect of passive image forensics in future.更多还原