【作者】 董莉莎；

【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】 随着数码相机的普及,数字图像以其方便、即时、易传输等优点,成为现代通信中一种重要的媒体形式。同时,由于简单易用的数字图像处理软件的不断发展与广泛流传,使得数字图像的伪造和篡改变得易如反掌。近年来,国内外在政治、司法、新闻等领域频繁出现的造假事件使得人们对于数字图像的真实性和完整性产生了质疑,因此数字图像取证技术,尤其是利用图像自身特性的盲取证技术,成为近年来学术界的研究热点。本文从重压缩操作对JPEG图像带来的统计特性上的影响为出发点,研究了JPEG重压缩操作的检测方法,以及利用重压缩检测来定位JPEG图像篡改区域的方法。本文的主要工作如下：(.1)提出了一种基于DCT系数灰度共生矩阵的重压缩检测方法。首先分析了重压缩操作对JPEG图像DCT系数的影响,考虑到之前的大部分重压缩检测算法都只考虑了DCT系数的一阶统计特性,而没有考虑相邻DCT系数间的相关性,这是导致之前算法检测准确率低的一个主要原因。本文从这一角度入手,分析了重压缩操作对DCT系数灰度共生矩阵的统计特性的影响,发现此二阶统计特征可以有效地放大重压缩操作给图像DCT系数带来的影响,进而提高重压缩检测的准确率。(2)提出了一种基于DCT系数首位有效数字Markov模型的重压缩检测算法。首先分析了重压缩操作对DCT系数首位有效数字分布特性的影响,并给出利用Markov过程来对DCT系数首位有效数字分布进行建模的方法。对比实验结果表明,该算法能够有效地检测重压缩操作,尤其是在第二次压缩质量远小于第一次压缩质量时,检测准确率较之前的算法有很大提高。(3)提出了一种基于重压缩检测的JPEG图像篡改定位模型。由于JPEG图像经过篡改后通常会再次保存为JPEG格式,因此会引入重压缩操作。分析表明,可以将篡改后的JPEG图像分为两部分,一部分为背景区域,这部分区域具有重压缩特性,而另一部分为篡改区域,这部分区域的重压缩特性大大减弱,基本不具有重压缩特性。因此,通过检测待测图像中的某一区域是否具有重压缩特性,可以定位图像中的篡改区域。实验表明,该模型能够有效检测和定位出图像的篡改区域,而且对旋转、缩放、羽化等操作具有鲁棒性,同时对不同压缩质量的篡改图像也具有鲁棒性。更多还原

【Abstract】 With the popularity of digital cameras, digital image has become one of the most important media forms in modern communication because of its convenience, instant and easy to transport. At the same time, with the development and widely spreads of the image processing software, tampering a digital image becomes an easy job for non-specialist and the tampered image is hardly detected by the naked eyes. In recent years, more and more tampered images emerged in the field of political, judicial, news, etc. both in domestic and overseas, which has caused skeptical about the integrity and authenticity of the images. Thus, image forensics, especially those technics using only the image characteristics, has become a hotspot of study in recent years. In this paper, considering double compression effects on the statistical characteristics of JPEG images, we proposed two different double compression detection methods, and a JPEG image tampering localization model using double compression detection features. The main contribution of this thesis includes:(1) Proposed a double compression detection method based on the Gray Level Co-occurrence Matrix (GLCM) of the DCT (Discrete Cosine Transform) coefficients. Through deep analysis about the double compression effects on the DCT coefficients of JPEG images and those methods proposed to detect double compression previously, we find that some correlations between adjacent coefficients, which have been ignored in the previous works, can be used to increase the detection accuracy. Based on this analysis, we find that the Gray Level Co-occurrence Matrix, which is a second order statistics, can reflect the correlations and thus enlarge the double compression effects on DCT coefficients. Experiments show that this second order statistics do increase the double compression detection accuracy compared to the first order statistics.(2) Proposed a double compression detection method based on the Markov model of the first digits of DCT coefficients. After a comprehensive analysis about the double compression effects on the distribution of the first digits of DCT coefficients, we proposed to model the distribution of the mode based first digits of DCT coefficients using Markov transition probability matrix and utilize its stationary distribution as features for double compression detection. Experiments show the effectiveness of the proposed method, especially when the second compression factor is much lower than that of the first one, the detection results have a significant improvement. (3) Proposed a JPEG image tampering localization Model based on double compression detection. Most of the tampered image will be resaved in JPEG format to save the storage space while maintaining the image quality, which may leave double compression artifacts. Study on the tampered JPEG image shows that it can be divided into two parts:one is the original part that has double compression artifacts and the other is the tampered part without double compression artifacts, thus by detecting whether the image blocks have double compression artifacts, the tampered region can be determined. Experiment results show that the proposed model is effective under post-processing operations such as rotation, resizing, feathering, etc. What’s more, the model gives promising results even when the tampered image has been compressed at a relatively low quality factor.更多还原