数字图像操作取证技术研究

发布时间：2018-08-02 18:57

【摘要】：随着数字媒体编辑技术的快速发展,数字图像修改或篡改变得非常容易,违规编辑和虚假图像也频繁出现,严重损害了数字媒体信息的可信性与安全性。在司法取证和新闻纪实等应用领域,常需要通过技术手段验证图像数据的原始性和真实性,并还原图像的操作历史以获取更多取证信息。因此,研究数字图像取证技术对于实现数字媒体内容认证具有重要意义和实用价值。本论文以数字图像操作取证技术为主要研究内容,取得创新性研究成果包括： 1)提出了两种基于边缘特征一致性分析的图像拼接检测方法。第一种方法利用自然边缘与拼接边界在CFA (Color Filter Array)内插特征一致性方面的差异,提出了一种快速有效的图像拼接检测算法,可有效克服现有大多数拼接检测技术需要统计学习导致算法复杂度较高的缺陷。第二种方法针对图像拼接后进行边界模糊润饰这一实际篡改情形,提出了一种基于边缘模糊度估计的拼接检测算法,可有效定位模糊润饰后的拼接边界。 2)提出了分别针对USM(Unsharp Masking)锐化和中值滤波的图像操作取证算法。在国际上较早提出了数字图像锐化和中值滤波取证问题。对图像USM锐化过程进行信号建模且给出了相应的数学描述,分析了过冲效应的产生机理,由此设计了有效的USM锐化检测方案。同时,从理论上分析了中值滤波所引起的图像一阶微分域统计特性异常,提出了一种快速有效的中值滤波操作检测算法。实验结果表明,此锐化和中值滤波取证算法均可有效鉴别相应的图像滤波操作。 3)提出了图像对比度增强取证系列算法。为有效检测中低等质量JPEG图像上的对比度增强操作,提出了一种基于灰度直方图峰谷形状分析的对比度增强检测方法；利用峰谷位置分布与像素值映射函数之间的对应关系设计了一种快速有效的伽玛参数盲估计方法；针对源图像区域经历不同的对比度增强情形,设计了一种有效的图像拼接检测方法；最后简要分析了现有对比度增强取证算法的安全性。实验结果表明,所提对比度增强取证相关算法均取得较高性能。 4)提出了一种半侵入式重采样算子源取证算法。从理论上推导出严格单调信号在经历传统型和几何抖动型重采样后一阶微分极性的变化规律,通过设计合适的模式图像,提出了一套完整的重采样算子源鉴别方法。实验结果表明,该方法既可识别重采样软件的内含算子,也可在特定情形下检测反取证型重采样操作。
[Abstract]:With the rapid development of digital media editing technology, digital image modification or tampering becomes very easy, illegal editing and false images appear frequently, which seriously damages the credibility and security of digital media information. In the field of judicial forensics and news documentary, it is often necessary to verify the originality and authenticity of image data by technical means, and to restore the operating history of image to obtain more evidential information. Therefore, the research of digital image forensics technology is of great significance and practical value for the realization of digital media content authentication. In this paper, digital image operation forensics technology is the main research content, and the innovative research results are as follows: 1) two image mosaic detection methods based on edge feature consistency analysis are proposed. In the first method, a fast and effective image mosaic detection algorithm is proposed by taking advantage of the differences between the natural edges and the stitched edges in the CFA (Color Filter Array) interpolation feature consistency. It can effectively overcome the high complexity of most existing stitching detection techniques due to the need of statistical learning. The second method is based on edge ambiguity estimation to solve the problem of edge fuzzy retouching after image stitching. It can effectively locate the stitching boundary after fuzzy retouching. 2) an image operation forensics algorithm for USM (Unsharp Masking) sharpening and median filtering is proposed. The problems of digital image sharpening and median filter forensics have been put forward earlier in the world. The signal modeling and mathematical description of image USM sharpening process are given. The mechanism of overshoot effect is analyzed and an effective USM sharpening detection scheme is designed. At the same time, the first order differential domain statistical anomaly caused by median filter is analyzed theoretically, and a fast and effective median filter operation detection algorithm is proposed. Experimental results show that both the sharpening and median filtering forensics algorithms can effectively identify the corresponding image filtering operations. 3) A series of image contrast enhancement forensics algorithms are proposed. In order to detect contrast enhancement in JPEG images of low and low quality, a contrast enhancement detection method based on peak and valley shape analysis of gray histogram is proposed. A fast and effective blind estimation method for gamma parameters is designed based on the correspondence relationship between peak and valley position distribution and pixel mapping function. An effective image mosaic detection method is designed, and the security of the existing contrast enhancement forensics algorithm is briefly analyzed. Experimental results show that the proposed contrast enhancement forensics correlation algorithms are of high performance. 4) A semi-invasive resampling operator source forensics algorithm is proposed. The variation law of the first order differential polarity of strictly monotone signals after the traditional and geometric jitter resampling is deduced theoretically. A set of complete resampling operator source identification method is proposed by designing suitable pattern images. The experimental results show that this method can not only identify the embedded operators of the resampling software, but also detect the anti-forensics resampling operations under certain circumstances.
【学位授予单位】：北京交通大学
【学位级别】：博士
【学位授予年份】：2013
【分类号】：TP391.41;D918.1

【参考文献】