改进的SIFT耦合特征点集群的图像伪造检测算法  

Image Forgery Detection Algorithm Based on Improved SIFT Coupled Feature Point Clustering

作　　者：柴建伟 刘婷 CHAI Jian-wei;LIU Ting(Langfang Yanjing Career Technical College,Langfang Hebei 065200,China)

机构地区：[1]廊坊燕京职业技术学院,河北廊坊065200

出　　处：《西南师范大学学报（自然科学版）》2018年第3期34-41,共8页Journal of Southwest China Normal University(Natural Science Edition)

基　　金：河北省自然科学基金项目(A2013203018);河北省重点科技攻关项目(12207107D-1)

摘　　要：针对当前图像伪造检测算法进行图像伪造检测时主要通过设定比例阀值来实现特征匹配,存在检测误差大、鲁棒性不强等不足,提出了改进的SIFT耦合特征点集群的图像伪造检测算法.首先,采用二进小波变换提取伪造图像的低频子带以用于特征点检测;然后,基于特征点邻域旋转不变纹理特性,改进了SIFT机制,生成新的特征描述子对其进行描述,减少误匹配,并提出了自适应匹配策略,通过搜索最优比例阀值,以提高算法检测精度及鲁棒性;最后,通过构建特征点的均值漂移向量,对特征点均值和特征点的偏差进行度量,实现特征点的集群,从而完成图像的伪造检测.仿真结果显示:跟当前的伪造检测方法相比,本文方法具有更高的检测精度与鲁棒性,呈现出较好的ROC特性.In view of the current image forgery detection algorithm for image forgery detection,the ratio of the threshold has mainly been set to achieve feature matching,which lead to the detection error is big;the robustness is not strong and so on.A effective image forgery detection algorithm based on improved SIFT coupled feature point clustering has been proposed in this paper.Firstly,the low frequency sub bands of the forged image are extracted by using dyadic wavelet transform to detect the feature points.Then,using feature point neighborhood rotation invariant texture characteristics to describe the feature points,formation characteristic descriptor to reduce the false matching,adaptive matching strategy is proposed,search the optimal ratio of the threshold in order to improve the accuracy and robustness of detecting algorithm,to improve the traditional SIFT feature descriptor generation and matching strategies.Finally,the mean shift vector of feature points is constructed to measure the deviation of feature points and feature points to realize the clustering of feature points and the image forgery detection is completed.The simulation results show that compared with the current forgery detection methods;this method has the characteristics of small detection error,strong robustness and so on.

关 键 词：图像伪造检测 SIFT 旋转不变纹理特性 均值漂移向量 特征描述子 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]