基于数字图像统计特性的篡改检测算法研究

【作者】 詹玲超；

【导师】 黄继风；

【作者基本信息】 上海师范大学 ， 计算机应用技术， 2007， 硕士

【摘要】 在过去的几十年里,随着计算机技术的进步,数字图像输入输出设备的发展和普及,以及功能强大的图像处理软件的开发,数字图像处理技术已经渗透到各个领域,制造出逼真的篡改图像是很容易的。如果这些篡改图像被犯罪分子滥用在媒体或法律上,将会对社会造成很大的影响,所以检测数字篡改已经成为一个非常严峻的问题。本文在对数字图像的一些统计特性进行研究的基础上,提出了两个检测篡改图像的算法。其中第一个是基于数码相机固有统计特性的检测算法,依据是数码相机传感器模板噪声的不变性和颜色滤波矩阵插值算法给图像带来的特殊相关性,先基于传感器的模板噪声初步检测出图像中的疑似篡改的区域,再通过检测由颜色滤波矩阵插值算法带来的相关性来进一步确认篡改区域。在篡改图像时,可能有很多不同的操作,但为了得到逼真的图像,模糊处理是必不可少的,所以篡改过的区域都是图像中较模糊的区域。文章提出的另一个算法就是通过检测图像中的模糊区域来检测篡改。实验结果表明,基于数码相机固有特性的检测算法能够精确地定位篡改区域,基于模糊估计的检测算法能够检测出图像中的模糊区域,即检测出图像本身的模糊和因篡改得到的模糊。将这两种算法应用于多幅篡改图像,都有很好的检测效果和性能。更多还原

【Abstract】 During the past decades, powerful computers, high-resolution digital cameras and sophisticated photo-editing software packages have become affordable and available to a large number of people. It has become fairly straightforward to create digital forgeries that are hard to distinguish from authentic photographs. These forgeries, if used in the mass media or courts of law, will have a great impact on our society. So the detecting of digital tampering has become a crucial problem.Two algorithms of detecting the digital forgeries are proposed in this paper. They are based on the study of some statistical properties of digital images. The first one is called detection of forgery using inherent attributes of the digital camera. It detects the forgeries on the basis of two inherent attributes of the digital camera, the sensor pattern noise’s invariability and the special correlation caused by color filter array interpolation algorithms. We detect the tampered regions with the sensor pattern noise, and then verify the result with the other inherent attribute.In the making of digital image frauds, there would be many manipulations. But applying blurring is inevitable to get vivid forgeries. As a result, the tampered regions are included in the blurred regions of the forgeries. Therefore, in this paper, we propose another algorithm to figure out the forgeries by detecting the blurred regions of images.Finally we implement the proposed algorithms and experiment on them. The result shows that the first algorithm can detect the tampered regions accurately and the other one can figure out the original blurred and tampered regions of images. Both of them are demonstrated on several forged images and have been proved to have high efficiency.更多还原