基于四元数极性复指数变换的图像复制-粘贴篡改检测算法

Image Copy-Move Tamper Detection Algorithm Based on Quaternion Polar Complex Exponential Transformation

在面对复制-粘贴(Copy-Move)篡改检测时,图像经过压缩、缩放、旋转和加噪等篡改操作处理后会导致检测性能的急剧下降甚至检测失败。本研究提出一种基于四元数极性复指数变换(quaternion polar complex exponential transform,QPCET)和改进PatchMatch的篡改检测算法。将待测图像进行重叠分块之后,计算各分块的四元数极性复指数变换不变量,并作为每一块的特征向量,构成特征矩阵。特征向量的初步匹配通过改进的PatchMatch算法完成,再通过随机抽样一致(random sampling consistency,RANSAC)算法去除错误的匹配块。在此基础上,使用形态学运算对标记区域进行处理,最终确定篡改区域。仿真结果表明,该算法可以有效检测图像中的Copy-Move篡改,并且对图像旋转、缩放、加噪和JPEG压缩等篡改区域后处理操作具有较好的鲁棒性。

In the face of Copy-Move tamper detection,post-tampering operations such as image compression,scaling,rotation,and noise addition can lead to a sharp decline in detection performance or even detection failure.This article proposes a tamper detection algorithm based on quaternion polar complex exponential transformation(QPCET)and improved PatchMatch.After overlapping and blocking the image to be tested,the quaternion polar complex exponential transform invariant of each block are calculated as the feature vectors of each block to form a feature matrix.The initial matching of feature vectors is completed through the improved PatchMatch algorithm,and then the erroneous matching blocks are removed through the random sampling consistency(RANSAC)algorithm.On this basis,morphological operations are used to process the marked area and ultimately determine the tampered area.The experimental simulation results show that this algorithm can effectively detect Copy-Move tampering in images.It also has good robustness to post-processing operations such as image rotation,scaling,noising,and JPEG compression.