摘要
针对当前较多图像伪造检测算法主要通过利用图像的灰度特征来进行图像伪造检测,当伪造内容存在较大灰度差异时,将导致检测结果中出现较多的误测以及鲁棒性不佳的不足。提出了基于区域直方图和特征相关匹配规则的图像复制-粘贴篡改检测算法。首先,利用Hessian矩阵行列式对图像特征进行检测,通过积分图像的方法提高Hessian矩阵行列式检测图像特征点的效率。然后,通过求取Haar小波响应值,以判定特征点的主方向。接着,对圆形窗口进行均匀分割,利用像素点的梯度模值来建立分割区域直方图,以获取特征向量,生成特征描述符。最后,利用归一化互相关(NCC)函数对特征点的相关性进行度量,建立特征相关匹配规则对特征点进行匹配。利用凝聚层次聚类方法,对特征点进行层次聚类,实现图像的伪造检测。实验结果分析表明,与当前图像匹配算法相比,图像伪造检测算法不仅能较精确的对伪造内容进行检测,而且还具有较强的鲁棒性能。
Aiming at the current image forgery detection algorithm mainly carried out image forgery detection using the image gray characteristic,when there is a big gray difference forged content which will lead to insufficient detection results in more robust error detection algorithm and poor performance. An algorithm for image copy-paste forgery detection based on region histogram and feature correlation matching rule is proposed in this paper. Firstly,the Hessian matrix determinant is used to detect the image features,and the efficiency of Hessian matrix determinant to detect image feature points is improved by integrating image. Then,the Haar wavelet response value is calculated in the circular window with the feature point as the center to determine the main direction of the feature points. Then,the circular window is divided evenly,and the histogram of the segmentation region is constructed by the gradient value of the pixel,and the feature vectors are acquired to complete the generation of the feature descriptor. Finally,the normalized crosscorrelation function is used to measure the correlation of feature points,and the feature correlation matching rules are established to match the feature points. Using agglomerative hierarchical clustering method,the hierarchical clustering of feature point is implemented to detect forgery. The experimental results show that this image forgery detection algorithm can not only detect the forgery content,but also have stronger robustness compared with the current image matching algorithm.
出处
《电子测量与仪器学报》
CSCD
北大核心
2018年第4期103-109,共7页
Journal of Electronic Measurement and Instrumentation
基金
国家自然科学基金(60073057)
河南省科技攻关计划(172102210117)
驻马店市科技计划(17135)资助项目
关键词
图像篡改检测
Hessian矩阵行列式
区域直方图
归一化互相关函数
特征相关匹配规则
凝聚层次聚类方法
image tampering detection
Hessian matrix determinant
region histogram
normalized cross correlation function
feature correlation matching rule
agglomerative hierarchical clustering method
