【作者】 鄂文；

【导师】 孔祥维；

【作者基本信息】 大连理工大学 ， 信号与信息处理， 2011， 硕士

【摘要】 照片作为生成证据的一种,在刑侦过程和司法体系中扮演着非常重要的角色。照片一般会以数字图像和打印图像两种形式递交法庭,在确认照片可以作为法庭的有效证据之前需要对其来源与真实性进行认证。以此为应用背景,图像取证领域的学者们一直把数字图像相机来源认证作为一个重要的研究方向,但是对于打印照片的相机来源认证却鲜有研究成果问世。事实上在很多时候是无法获得原始数字图像的,例如照片的提供者只保留了数码照片的打印版本而删除了原始数字图像,或者提供的数字图像本身就是经过扫描获取到的等等。因此,如何能够确保这类图像来源的真实性和可靠性,也成为了当下维护社会稳定和司法公正急切需要解决的新问题。本文将对打印照片的相机来源取证技术展开研究,主要工作总结如下：(1)提出了一种基于模式噪声的打印扫描图像来源取证方法。本文首先对传统的数字图像相机来源取证技术进行了研究,并对相机的成像系统以及在此过程中成像器件和处理操作对生成的数字图像产生的的影响进行了介绍。对于在数字图像来源取证中对不同相机来源的区分度较高的相机传感器模式噪声,分析了其作为特征进行打印扫描图像相机来源检测的可行性,通过比较选择出一种能够较大程度上滤除不含原始图像信息成分的滤波器用于提取噪声。最后,针对图像二次获取过程中可能引入的几何失真问题,提出了基于相机传感器模式噪声的打印扫描图像来源取证方法。实验结果表明,该方法对于打印扫描图像进行来源检测的平均准确率达到了76.8%,比Fridrich的方法提高了5%。而且该方法对于图像二次获取过程中引入的旋转、缩放、剪切等影响,均能达到稳定的检测效果。(2)提出了一种基于高光成分去除的打印扫描图像来源取证方法。通过对前面提出方法在实验中失效图像与有效图像的比较,发现扫描过程中引入的高光可能会对检测的结果产生不利影响。为了验证这一假设并能更加准确地检测图像来源,本文从高光产生的机理入手,利用双色反射模型和镜面反射-漫反射机制将图像中的镜面反射成分和漫反射成分进行了分离,从而达到了去除高光的目的。通过对四种品牌的相机进行的对比实验证明该方法能够将打印扫描图像的初始相机来源检测的平均准确率提高到84.3%。更多还原

【Abstract】 As a kind of generated evidence, photo becomes more and more important in the judicial system and the criminal investigation. It usually has two forms, digital image and printed image, when delivered to court. Before the court receives the photo as evidence, its authenticity and source need to be identified. For this application background, scholars in the field of image forensics have been considered the digital image camera source identifying as an important research direction. But for printed photos, there are few findings come out.In fact, we cannot obtain the original digital image in many cases. For example, provider only has the print version of the digital image with the original image deleted or the digital image delivered it is a recaptured image. Therefore, how to ensure the authenticity of these images and the reliability of the source becomes a new problem need to address for the urgent justice and social stability. In this paper, we will research the print-scanned image forensics technology, and main works are summarized as follows:(1) Proposed a print-scanned image source identification method based on Pattern Noise.In this paper, the traditional source of digital image forensics technology has been studied first. Then we introduced the imaging system of camera and the influences bring to the generated image by imaging device and processing operations in detail. Camera sensor Pattern Noise has higher differentiation of different cameras in digital image sources forensics. Through the comparison, we choose a filter which can ensure remaining the least information of original image to extract noise. At last, for the geometric distortion introduced by print-scan process, we proposed a print-scanned image source identification method based on Pattern Noise. Using this method, the average identification accuracy rate of test images reaches to 76.8%, which improves 5% compared with Fridrich’s method. And the method can also reach the stable test results against the rotation, scale and cut introduced by the recapturing process.(2) Proposed a source camera identification method of print-scanned digital image based on highlight removement.Through the comparison and the analysis of the experiment before, we find that the failure images usually have visible highlight which could be the reason caused the lower result. In order to verify this hypothesis and solve the problems found in experiments, we analyze the generation of highlight. Using DR model and Specular-diffuse reflection, we separate specular reflection components and diffuse components of the image. Experiment results show that this is an effective method for source camera identification with less restrict, the average accuracy on distinguishing four cameras of print-scanned images improved to 84.3%更多还原