基于图像变换和支持向量机的数字图像拼接检测技术研究

发布时间：2018-07-20 15:02

【摘要】：随着科学技术的蓬勃发展,越来越多功能强大的图像编辑软件在我们生活中普及开来,这样导致数字图像越来越容易被人为编辑和篡改,并且不留任何明显的痕迹就可达到视觉上的逼真。当篡改后的数字图像再被应用于医保报销、警务系统和法庭取证等方面时,其不再具有使用价值;虚假的图像会对国家、社会产生巨大的危害,造成无法估量的损失。所以,数字图像篡改取证技术成为当下一个非常重要的研究课题。图像篡改、伪造最常见的手段之一便是拼接操作,本文主要探讨拼接篡改的被动取证技术,通过对大数量已知真假的图像进行分析,找出这些图像在拼接篡改操作过程中一些敏感特征的变化规律,然后利用图像变换和支持向量机进行拼接检测。主要内容及成果如下:1.阐述了数字图像取证技术的研究背景与意义,当前常见的数字图像篡改、伪造手段,以及数字图像取证技术的国内外研究现状;2.介绍了数字图像拼接取证基本理论和相关技术,包括图像的具体拼接过程、拼接被动取证系统的基本框架、取证时涉及的一些图像变换、支持向量机理论,最后介绍了一些典型拼接被动取证技术的具体方法;3.提出一种结合DTT变换、DCT变换和最小二乘支持向量机的图像拼接检测算法。算法首先对图像分别做DTT变换和DCT变换;然后计算变换后DTT域和DCT域的马尔科夫状态转移概率矩阵;将状态转移概率矩阵作为特征向量送入最小二乘支持向量机训练以获得预测模型,依据该模型判别待测图像是否经过拼接篡改;最后分别在中国科学院自动化所篡改图像检测评估库和哥伦比亚大学图像拼接检测评估库进行实验,与传统算法做对比,实验结果表明所提算法拥有更高的拼接检测准确率。4.提出一种基于对偶数复小波变换和最小二乘孪生支持向量机的图像拼接检测算法。算法首先将待测图像进行对偶数复小波变换以获得不同的子带图像,然后计算子带图像的马尔科夫状态转移概率矩阵,将该概率矩阵作为拼接特征向量送入最小二乘孪生支持向量机训练以获得预测模型,依据该模型判断待测图像是否经过拼接篡改。在哥伦比亚大学无压缩图像拼接检测评估库和哥伦比亚大学图像拼接检测评估库上分别进行实验,与传统算法做对比,实验结果证明本章所提算法具有更高的拼接检测准确率。
[Abstract]:With the rapid development of science and technology, more and more powerful image editing software is popularized in our daily life, which makes digital images more and more easy to edit and tamper with. And can achieve visual lifelike without leaving any obvious trace. When the tampered digital image is applied to medical insurance reimbursement, police system and court evidence, it no longer has the use value; false images will cause great harm to the country and society, resulting in incalculable losses. Therefore, digital image tampering and forensics technology has become a very important research topic. One of the most common methods of image tampering and forgery is stitching. This paper mainly discusses the passive forensics technology of mosaic tampering, and analyzes a large number of known true and false images. The changes of some sensitive features of these images in the process of mosaic tampering are found out, and then the image transformation and support vector machine are used for stitching detection. The main contents and results are as follows: 1. This paper expounds the research background and significance of digital image forensics, the common methods of digital image tampering, forgery, and the research status of digital image forensics both at home and abroad. This paper introduces the basic theory and related technology of digital image stitching, including the process of image stitching, the basic frame of passive forensics system, some image transformation and the theory of support vector machine. In the end, some typical methods of passive forensics are introduced. An image mosaic detection algorithm based on DTT transform and least squares support vector machine (LS-SVM) is proposed. Firstly, DTT transform and DCT transform are performed on the image, then the Markov state transition probability matrix in DTT domain and DCT domain is calculated. The state transition probability matrix is fed into the least squares support vector machine (LS-SVM) as the eigenvector to obtain the prediction model, according to which the image to be tested is judged whether the image is tampered with or not. Finally, the experiments were carried out in the tampered image detection and evaluation database of Chinese Academy of Sciences and Columbia University image mosaic evaluation database, and compared with the traditional algorithm. The experimental results show that the proposed algorithm has a higher accuracy rate of stitching detection. 4. An image mosaic detection algorithm based on dual complex wavelet transform and least square twin support vector machine is proposed. The algorithm firstly performs even number complex wavelet transform to obtain different sub-band images, and then calculates Markov state transition probability matrix of sub-band images. The probability matrix is put into the least square twinning support vector machine as the splicing feature vector to obtain the prediction model. The model is used to determine whether the image under test has been spliced and tampered with. Experiments were carried out on Columbia University uncompressed image mosaic evaluation database and Columbia University image mosaic detection evaluation database respectively. The experimental results show that the proposed algorithm has a higher detection accuracy compared with the traditional algorithm.
【学位授予单位】：安徽大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【相似文献】