您的位置:首页  > 论文页面

基于SIFT的图像复制粘贴篡改检测方法

发表时间:2017-11-30  浏览量:996  下载量:114
全部作者: 杨培,章登勇,夏明,杨高波
作者单位: 中国湖南大学信息科学与工程学院
摘 要: 图像在进行复制粘贴篡改操作时,往往经过缩放、旋转等操作以迷惑观察者,且复制粘贴篡改区域有时会距离相对较近。提出一种基于尺度不变特征转换(scale-invariant feature transform,SIFT)关键点的篡改检测改进算法。首先,采用SIFT 描述符提取图像的SIFT 关键点,采用递归折半的广义2 最近邻(generalized 2 nearest neighbor,g2NN)匹配策略对关键点进行匹配;然后,利用一种基于匹配向量方向投票的策略对关键点进行分类;随后,通过随机抽样一致(random sample consensus,RANSAC)算法和二次聚类降低误检,得到仿射变换参数;最后,根据仿射变换模型,使用零均值归一化互相关(zero-mean normalized cross correlation,ZNCC)准则衡量待测图与仿射变换后的变换图之间的相关性以定位篡改区域。实验结果表明,该算法具有较高的检测效率和良好的定位效果。
关 键 词: 数据安全与计算机安全;复制粘贴篡改检测;匹配向量方向投票;仿射变换;零均值归一化互相关
Title: A SIFT-based image copy-move forgery detection approach
Author: YANG Pei, ZHANG Dengyong, XIA Ming, YANG Gaobo
Organization: College of Computer Science and Electronic Engineering, Hunan University
Abstract: Copy-move forgery detection is often attacked by geometric transformation (such as scaling, rotation) to confuse observer. The distance between original region and tampered region is sometimes very close. This paper presents an improved scale-invariant feature transform (SIFT)-based image forensic method for copy-move forgery detection. First, SIFT key points are extracted by descriptor and matched by recursive bisection generalized 2 nearest neighbor (g2NN). Then, the matched key points are classified by matched vector direction voting strategy. Mismatch is decreased by random sample consensus (RANSAC) and re-clustering. After obtaining affine transform parameters, candidate image is transformed by affine model. The correlation map between test image and transformed image is estimated by zero-mean normalized cross correlation (ZNCC) strategy, and the tampered region is located based on correlation. Experimental results show that the proposed approach achieves high detection efficiency and desirable tampering location.
Key words: data security and computer security; copy-move forgery detection; matched vector direction voting; affine transform; zero-mean normalized cross correlation
发表期数: 2017年11月第22期
引用格式: 杨培,章登勇,夏明,等. 基于SIFT的图像复制粘贴篡改检测方法[J]. 中国科技论文在线精品论文,2017,10(22):2459-2466.
 
0 评论数 0
暂无评论
友情链接