基于能量约束与结构相似聚类的图像篡改检测  

Image tampering detection algorithm based on energy constrained coupled structure similar clustering

作　　者：王晓雨[1] WANG Xiaoyu(College of Computer Engineering,Jingchu Institute of Technology,Jingmen Hubei 448000,China)

机构地区：[1]荆楚理工学院计算机工程学院,湖北荆门448000

出　　处：《太赫兹科学与电子信息学报》2021年第3期478-484,共7页Journal of Terahertz Science and Electronic Information Technology

基　　金：湖北省教育厅科研计划研究项目(B2020192);荆门市科技局科研项目(2019YDKY078)。

摘　　要：借助能量约束与结构相似聚类机制,设计了一种新的图像内容伪造检测算法。首先,借助Hessian算子,利用盒式滤波器来生成Hessian行列式,以准确检测图像特征。然后,通过计算图像的Haar小波值,求取图像的方向信息,以构建图像特征的邻域窗口。再计算该邻域窗口内像素点的曲率信息,构成鲁棒性较好的特征向量。最后,对图像特征进行欧氏距离度量,并联合图像的区域能量特征,完成度量结果的约束,以实现图像特征的精确匹配。采用结构相似度(SSIM)函数,聚类匹配结果识别伪造区域,实现准确的检测。仿真数据表明,较当前内容检测技术而言,在多种几何变换干扰下,本文算法具有更高的检测准确性与鲁棒性。A new image content forgery detection algorithm by means of energy constraints and structure similarity clustering mechanism is proposed.Firstly,with the help of Hessian operator,a box filter is utilized to generate row Hessian formulation to accurately detect image features.Then,by calculating the Haar wavelet value of the image,the direction information of the image is obtained to construct the neighborhood window of the image features.Then the curvature information of the pixels in the neighborhood window is calculated to form a robust eigenvector.Finally,Euclidean distance measurement is adopted to measure image features,and regional energy features are combined to complete the constraints of measurement results,so as to achieve accurate matching of image features.Structural Similarity Index(SSIM)function is utilized to identify forgery areas by clustering matching results,and accurate detection is realized.The simulation results show that,compared with current content detection technology,the proposed algorithm has higher detection accuracy and robustness under the interference of various geometric transformations.

关 键 词：图像复制-粘贴篡改检测 Hessian算子 曲率信息 能量约束 结构相似聚类 

分 类 号：TN911.73[电子电信—通信与信息系统]