【作者】 曹雁军；

【导师】 高铁杠；

【作者基本信息】 南开大学 ， 计算机应用技术， 2013， 博士

【摘要】 当今信息时代,随着计算机、网络技术、多媒体技术以及廉价数码设备的广泛使用,人们可以更加方便、快捷的获取一幅数字图像,在人们享受数字图像所带来快乐的同时,功能强大的图像处理及编辑软件应运而起,它们的出现和不断完善使得一些怀有不良目的的人可以轻而易举地对一幅数字图像作品进行恶意篡改,以此来掩饰图像的真实面貌,从而达到某种特定目的。尤其是当被篡改后的图像应用到新闻媒体、司法以及军事等领域时,势必会影响新闻和娱乐信息的真实性、司法审判的公正性、军事情报的可靠性,无论于国于民都会产生不可估量的损失。因此研究鉴别数字图像内容真实性的新技术,无论对重树人们对新闻媒体的信心,还是对维护社会公平和正义都具有深远的意义。本文在国家自然基金项目和天津市重点自然基金项目的支持下,重点围绕鉴别数字图像真伪这一主线,从图像内容篡改盲检测以及反取证技术两方面分别进行了研究,其主要研究成果如下：1.针对当前区域复制一粘贴篡改检测方法所提取的特征维数过大缺点,提出了一种基于DCT系数能量分布的篡改检测算法。将图像从空域转化到频域,对图像进行分块,利用离散余弦变换及其变换后的离散余弦系数的低频、中频、高频能量分布情况,依次对每个子块提取4维匹配特征,保证了算法具有较低的复杂度。并通过实验对所提方法在检测精度、高斯白噪声、高斯模糊以及多处篡改操作上进行性能分析,说明了所提算法具有较高的鲁棒性。2.提出了一种能抵抗旋转操作的图像盲取证方法,通过对图像统计特征的分析,提出了利用伪Zernike矩作为图像特征,以此来抵抗图像篡改中的旋转操作。将原始图像进行划分,利用低阶伪Zernike矩代表图像轮廓,高阶伪Zernike矩代表图像细节的特点,对每个子块构造低阶伪Zernike矩,提高了算法对噪声的鲁棒性,并且在构造伪Zernike矩的过程中采用递归方式,使算法复杂度由原来的(N2p3)降到0(N2p2),’极大的提高了算法的运行效率。3.提出了一利t能定位篡改区域并且能估计篡改区域旋转角度及缩放因_子的盲检测方法。将图像的空域和频域相结合,对图像进行划分并将每个区域从空域坐标映射为对数极坐标,保证了在空域坐标中经过旋转、缩放的篡改区域,在对数极坐标下仅仅为沿极轴的平移操作。采用PCA技术去除特征中的冗余数据,利用傅里叶变换以及相位相关频谱来对篡改区域进行定位,并首次根据相位相关频谱来估算篡改区域的旋转角度和缩放因子。大量实验结果表明,该方法能够准确的检测出经过旋转、缩放的篡改区域,并且对图像扭曲操作和图像增强操作也有良好的鲁棒性。4.针对现有的单次JPEG压缩取证技术,首次将混沌理论应用到反取证技术当中,提出了利用混沌噪声来去除JPEG压缩痕迹。首先利用压缩后的DCT系数间的方差来估计JPEG图像未压缩前各个DCT系数子带的分布模型,根据所估计模型,对于各个DCT系数子带,分别采用Logistic映射产生混沌噪声,并将混沌噪声添加到相应子带中。通过大量实验分析表明,通过添加混沌噪声,可以成功的消除JPEG压缩所带来的DCT系数直方图中的“沟壑“现象,并且该方法可以成功的欺骗现有的基于检测JPEG图像块效应的取证技术。比起同类技术,在一定的压缩因子下,该方法可以保证在添加混沌噪声后,图像有较小的视觉失真。最后,论文对本文工作进行总结,并给出下一步工作中的研究方向。更多还原

【Abstract】 Nowadays, with the widely use of the computer network、multimedia technology and low cost digital equipment, it is more convenient to access a digital image. However, when people enjoying the entertainment brought by digital images, the powerful image editing software arose that make it easier for those evil people to modify a digital image. Obviously, when the tampered image was used by the news、 judicature and military, it may mislead the public and throw doubt on the authenticity of the multimedia image, the fairness of the judicial evidence and the reliability of the military intelligence. So, the research of the identification technology for digital image is significative not only to rebuild people’s confidence in the news media, but also to maintain social fariness and justice.Based on the projects granted by the National Nature Science Foundation and Key Program of Natural Science Fund of Tianjin, we study on the identification technology for digital image and pay particular attention to the image content detection and anti-forensic technique. The main contributions of this dissertation are as following:1. A scheme for detecting the tampered regions based on the energy distribution of the DCT coefficients was proposed. Firstly, the image was transformed from spatial domain to frequency domain, then it was split to sub-blocks and each sub-block was represented by the transformed DCT coefficients, according to the nature of the Discrete Cosine Transform, we extracted only four features for each block, which significantly reduce the computational complexity. The experiments show that the proposed scheme could locate the doctored regions accurately, moreover, it was also robust to Gaussian white noise, Gaussian blur and multiple tampered operations.2. A robust scheme for identifying the doctored areas that underwent rotation was proposed. Through the analysis of the image statistical characteristics, in order to resist the rotation operation, the Pseudo-Zernike moment was used as the image feature. The image was divided into overlapped blocks, and then n-dimensional low order Pseudo-Zernike moments were extracted for each block, which ganrantee the proposed scheme is more robust to noise adding. During the feaure extraction, a recursive method is adopt, which reduce the computional complexity from O(N2p3) to O(N2p2).3. A passive detection scheme that can not only locate the tampered regions but also estimate the rotation angle and scaling factor was proposed. The image was segmented into a series of small blocks, each block was mapped to the log-polar domain, which can assure the rotated or scaled tampered areas were just represented by the translation along the log-polar coordinate. In the matching step, PCA was applied to make dimension reduction and the tampered regions were located by using the Fourier transform and phase correlation technique, moreover, by using the phase correlation spectrum, the proposed scheme could estimate the rotation or scaling parameter of the tampered areas perfectly. The experimental results show that the proposed scheme is also robust to image distortion and enhancement operation.4. A new anti-forensic scheme for the single JPEG compression image was proposed. The chaos theory was applied to the anti-forensic method for the first time. Firstly, the original JPEG image was transformed by the discrete cosine transform, and the model for each DCT sub-band was estimated according to the transformed DCT coefficients, then by using the chaos technique and the estimated model, the chaotic noise was added to each DCT sub-band. The experiments show that, by adding chaotic noise, the proposed scheme could erase the "gully" which caused by the JPEG compression perfectly, moreover, the proposed scheme could also deceive the existing forensic method that based on detecting the JPEG block artifact, compared with the competing method, under certain compression factor, the proposed method only introduce little visual distortion when adding chaotic noise.Finally, the paper makes a summary of the work, and points out the new research direction for the future work.更多还原