基于RG-SIFT的图像复制粘贴篡改取证算法

Image Copy-Paste Forensics Method Based on RG-SIFT

在以审判为中心背景下的刑事诉讼制度改革中,图像证据的审查与采信尤为重要。数字图像篡改操作中最为常见的就是复制粘贴,通过复制图像中的一块区域粘贴到其他部分,以达到增加或掩盖部分图像内容,同时伴随着几何变换以增加真实性。当前存在的检测类似篡改的算法大多数只能应对某种几何变换,计算复杂性较高且不能应对多区域复制粘贴的情况。为此,提出一种利用颜色和纹理信息提取尺度不变特征的方法,通过检测图像颜色通道下的局部不变性特征提取关键点,并使用层次聚类算法进行聚类,对关键点匹配结果进行仿射变换估计,最后使用随机抽样一致算法去除无匹配关键点。实验结果表明,该算法能够较好的处理经过多种变换的多区域复制粘贴篡改检测,能够在公安图像检验鉴定工作中发挥作用。

In the reform of criminal procedure system centered on trial, the examination and acceptance of image evidence is particularly important. The most common operation of digital image tampering is copy and paste, by copying a region of the image paste to other parts, in order to increase or cover up part of the image content, accompanied by geometric transformation to increase the authenticity. Most of the existing algorithms for detecting similar forgeries can only deal with a certain geometric transformation, which has high computational complexity and cannot deal with the situation of multi-region copy and paste. A method of extracting scale-invariant features based on color and texture information is proposed. At first, the key points are extracted by detecting the local invariant features in the color channel of the image, and the hierarchical clustering algorithm is used for clustering. And then, the affine transformation estimation is carried out on the matching results of the key points. At last, the random sampling consensus algorithm is used to remove the unmatched key points. Experimental results show that the algorithm can better deal with the multi-region copy-paste tamper detection after a variety of transformations, and can play a role in the public security image inspection and identification work.