摘要
当前较多的图像伪造检测算法直接将未经优化的图像特征提取结果用于篡改检测,忽略了伪图像特征的影响,导致其不能较为准确地检测伪造内容。对此,本研究提出了一种优化因子耦合比值制约规则的方法,用于检测伪造图像。首先,引入高斯差分(Difference of Gaussian,DoG)算子,通过比较像素点与其邻域点像素值的方法来提取图像特征。再利用DoG算子的二次Taylor展开式,构造优化因子,从提取的图像特征中筛选出不稳定的伪特征,以提高伪造内容的检测准确度。随后,在图像特征点的极坐标系中,建立特征点的8邻域区,用以计算特征向量。然后,通过特征向量求取特征点间的欧式距离信息,以及通过归一化互相关(Normalizedcrosscorrelation,NCC)函数求取特征点间的互相关信息,从而完成图像特征的匹配。最后,借助特征点间欧氏距离信息构造比值制约规则,对匹配特征点进行归类,以识别篡改内容。实验结果表明,较当前图像篡改检测方法而言,在各种几何变换操作下,本研究提出的算法具有更高的检测精度与鲁棒性,能够较为完整地检测出篡改内容。
In many current image forgery detection algorithms,the unoptimized image feature extraction results are directly applied to tamper detection,ignoring the influence of false image features,which makes the algorithm unable to detect the forgery content more accurately.To solve this problem,an image copy-paste tamper detection algorithm based on optimization factor coupling ratio restriction rule is proposed.Firstly,the Gauss difference operator is introduced to extract image features by comparing the pixel values of the pixels and their neighboring points.By using the second Taylor expansion of Gauss difference operator,an optimization factor is constructed to screen out unstable pseudo-image features from the extracted image features,so as to improve the detection accuracy of the algorithm.Next,in the polar coordinate system of the image feature points,eightneighborhood regions of the feature points are established.By calculating the Haar wavelet values of the neighborhood regions,the feature vectors are formed.Then,the Euclidean distance information between feature points is obtained by eigenvector,and the cross-correlation information between feature points is obtained by normalized cross-correlation model.The Euclidean distance information between feature points and cross-correlation information are used to match image features.Finally,on the basis of Euclidean distance information between feature points,ratio restriction rules are constructed to classify matching feature points to identify tampered content and form detection results.The experimental results show that the proposed algorithm can detect tamper content more completely,and the detection accuracy is higher than the current tamper detection algorithm.
作者
杜媛
DU Yuan(School of Big Data Application,Xi'an Vocational and Technical College,Xi'an 710072,China)
出处
《系统仿真技术》
2023年第2期148-155,共8页
System Simulation Technology
基金
陕西职业教育乡村振兴研究院课题(22YB022)
关键词
复制-粘贴篡改检测
优化因子
高斯差分算子
伪图像特征
极坐标系
比值制约规则
copy-paste tampering detection
optimization factor
Gauss difference operator
pseudo image feature
polar coordinate system
ratio restriction rule
