基于向量点积与相似聚类的图像伪造检测算法  

Image forgery detection algorithm based on vector dot product coupled similarity clustering

作　　者：刘琴琴 邱建林[2] LIU Qin-qin1, QIU Jian-lin2(1. College of Computer and Information Engineering, Nantong Institute of Technolgy, Nantong 226000,China;2. College of Computer, Nantong University, Nantong 226000, Chin)

机构地区：[1]南通理工学院计算机与信息工程学院,江苏南通226000 [2]南通大学计算机学院,江苏南通226000

出　　处：《计算机工程与设计》2018年第6期1685-1690,共6页Computer Engineering and Design

基　　金：国家自然科学基金项目(61272424);计算机软件新技术国家重点实验室开放课题基金项目(KFKT2012B29);江苏省自然科学基金项目(BK2010277)

摘　　要：为提高图像伪造内容的检测精度,对向量点积耦合相似聚类的图像伪造检测算法进行研究。利用Forstner检测算子提取图像的特征点,将特征点作为中心,建立不同步长的同心圆区域,以30°角为步长构建角度盘,求取梯度累计直方图,改进SURF生成特征向量的过程,输出特征描述符;求取特征描述符之间的余弦,形成向量点积,构造双阀值匹配机制,完成特征点的匹配;利用归一化互相关函数,度量特征点的相似性,根据其相似度完成特征点的聚类。仿真分析结果表明,与当前图像伪造检测算法相比,所提算法具有更高的检测效率与精度。To improve the detection accuracy of image forgery content,the image forgery detection algorithm based on vector dot product and similarity clustering was proposed.The feature points of the image were extracted accurately using Forstner operator,the feature points were taken as the center to construct concentric circles with different step sizes,30 degree angle was used to construct the angle plate for computing the gradient histogram,and the process of generating feature vectors was improved to obtain feature descriptors.The cosine between the feature descriptors was obtained to form a vector dot product for constructing a double threshold matching mechanism to complete the matching of feature points.The normalized cross-correlation function was used to measure the matching feature points for clustering the feature points.Simulation results show that the proposed algorithm has higher detection efficiency and accuracy compared with the current image forgery detection algorithm.

关 键 词：图像伪造检测 向量点积 相似聚类 Forstner检测算子 角度盘 双阀值匹配 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]