一种鲁棒的图像复制粘贴篡改盲检测方法  

A Robust Copy-move Forgery Blind Detection Method for Digital Image

作　　者：张墩利 周国栋 Zhang Dunli;Zhou Guodong(College of Intelligent Manufacturing College,Hunan Open University,Changsha 410004,China)

机构地区：[1]湖南开放大学智能制造学院,长沙410004

出　　处：《机电工程技术》2023年第9期137-140,214,共5页Mechanical & Electrical Engineering Technology

基　　金：湖南省教育厅2022年科学研究项目(22C0971);湖南省教育厅2022年科学研究项目(22C0972);湖南开放大学2022年科研课题(XDK-2022-C-2)。

摘　　要：针对现有图像复制粘贴篡改盲检测技术鲁棒性不强,容易受图像压缩、高斯噪声、亮度和对比度调整等后处理干扰的问题。提出一种新的基于图像块的高效检测方法,仅利用离散余弦变换矩阵的低频系数符号,提取图像块的特征向量;通过改进的元胞自动机规则对特征向量进行演化,有效将数据量从32维降至5维。再利用KD树模型获取每个图像块的5个最近邻相似目标,最后通过自定义后处理规则剔除假阳性判定结果。实验结果表明,和同类方法相比,所提方法即使在强后处理干扰环境下,依然能表现出更强的鲁棒性,同时还兼具较好的处理速度。在压缩和高斯噪声干扰上,检测精度平均提高了15%;而在亮度和对比度调整干扰上,检测精度平均提高了6%。It’s not so capable in disturbance rejection for current blind copy-move forgery detection methods(CMFD),such as image compression,Gaussian noise,brightness and contrast adjustment.A new robust block-based detection method is proposed.The feature vector is extracted from each block by only sign information of low frequency coefficients which come from discrete cosine transform(DCT)matrix of the image.Cellular Automata is employed to reduce the feature vector from 32 dimensions to 5 dimensions.Feature vectors are matched by utilizing the KD-tree based on the nearest-neighbor searching method to find the duplicated areas.Finally,the false matches are filtered and the forgeries are visualized inside the image.Experimental results show that the proposed method performed exceptionally well relative to the other state-of-the-art methods even when an image is heavily affected by the post-processing attacks.At the same time,it also shows the faster processing speed.The precision is improved by 15%on average under compression and Gaussian noise attacks,and 6%is improved under the brightness and contrast change attacks.

关 键 词：图像取证 复制粘贴篡改 元胞自动机 离散余弦变换 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]