基于Harris与LCP的同图复制粘贴篡改检测

Detection of copy-paste image tampering in same image based on Harris and LCP

在声像资料鉴定工作中,利用同图复制粘贴进行图像篡改的案例较为常见。针对此种篡改方式,提出一种基于特征点和特征值进行匹配的检测方法。首先利用Harris角点检测器进行特征点提取,以特征点为中心建立一个方形邻域,对方形邻域用局部配置模式方法(Local Configuration Pattern,LCP)提取特征,之后对特征向量进行字典排序,通过计算相邻特征向量的相似度确定匹配特征向量,最后利用RANSAC算子消除错误匹配点。实验结果表明,此法可有效检测图像篡改区域,对旋转、亮度、模糊、噪声等多种复杂操作均具有较好的鲁棒性,并能实现图像多区域篡改方式的检测。

In the identification of audio-visual resources,it’s common that there are some tampered cases by copy-paste.For this tampering method,this paper proposes a detection method based on the matching of feature points and eigenvalues.Firstly,the Harris point detector is used to extract feature points,around which a square field is built and the features are extracted by local configuration pattern(LCP).The feature vectors are sorted by dictionary,and the similarities of adjacent vectors are calculated to determine the matching feature vectors.Finally,using RANSAC operator to eliminate mismatch points.The experimental results show that this method can effectively detect tampering regions,with good robustness to complexities such as rotation,brightness,blur,noise,and can successfully detect multiregion tampering.