曲率尺度空间耦合旋转约束的图像篡改检测  

An Image Tampering Detection Algorithm Based on Curvature Scale Space Coupled Rotation Constraint

作　　者：李明妍 徐永刚[1] 刘英男[3] LI Ming-yan;XU Yong-gang;LIU Ying-nan(China University of Mining and Technology,Xuzhou Jiangsu 221116,China;Xuzhou Finance and Economics Branch of Jiangsu United Vocational College,Xuzhou Jiangsu 221008,China;Shanghai University of Engineering and Technology,Shanghai 201620,China)

机构地区：[1]中国矿业大学,江苏徐州221116 [2]江苏联合职业技术学院徐州财经分院,江苏徐州221008 [3]上海工程技术大学,上海201620

出　　处：《计算机仿真》2022年第1期166-171,176,共7页Computer Simulation

基　　金：国家自然科学基金项目(51574232,51204186);江苏省自然科学基金(BD2012129)。

摘　　要：针对当前复制-粘贴篡改检测算法存在的识别准确度不高的问题,提出了曲率尺度空间耦合旋转约束聚类的图像篡改检测方法。借助Canny算子,计算图像边缘曲线,再引入曲率尺度空间特征提取方法,通过像素点与高斯函数的卷积运算,求取曲线曲率,从而获取图像的初始特征点。采用初始特征点的二次Taylor项,对其真伪性进行辨别,以精确提取图像特征。并在特征点FFT(·)小波的基础上,求取特征向量。采用特征向量对图像特征进行距离测量,利用图像特征间的距离信息作为其依据,对图像特征进行准确匹配。通过旋转图像及匹配特征点来获取其旋转角度,利用该旋转角度作为约束条件,对匹配的图像特征进行聚类,以定位出篡改内容。实验结果显示,较现有的篡改经检测算法而言,在多种不同的几何变换下,所提方法的检测准确度更高。所提算法对多种内容篡改均有较高的检测精度与稳健性,能给信息安全领域提供较好的技术依据。Aiming at the defect as low recognition accuracy of current copy paste tamper detection algorithm, this paper proposes a tamper detection method based on curvature scale spatial feature extraction. With the help of Canny operator, the edge curve of the image was calculated, and the curvature scale space feature extraction method was introduced to calculate the curvature of the curve for obtaining the initial feature points of the image by convolution of the pixel and Gaussian function. The second Taylor term of the initial feature points was used to distinguish the authenticity of the initial feature points for extracting the image features accurately. And on the basis of Haar wavelet, the feature vector was obtained. The feature vector was used to measure the distance of image features, and the distance information between image features was used as the matching basis to accurately match image features. The rotation angle was obtained by rotating the image and its matching feature points, and the rotation angle was used as the constraint condition to cluster the matching image features for locking the tampering content. The experimental results show that the proposed algorithm has higher detection accuracy than the existing algorithm under many different geometric transformations. This algorithm has a high detection accuracy and robustness for a variety of content tampering, which can provide a better technical basis for the field of information security.

关 键 词：图像篡改检测 曲率尺度空间 旋转约束 曲线曲率 特征聚类 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]