基于篡改区域轮廓的图像拼接篡改盲取证算法

为了解决传统图像拼接篡改检测方法特征单一、检测方法固化、篡改区域定位偏差大等问题,提出了一种基于篡改轮廓的图像拼接篡改检测方法。该方法对图像拼接篡改痕迹进行分析,在频域中建模,提取多步马尔可夫特征表征系数转移特性,并提取局部信息熵与差分激励表征当前位置系数分布变化情况。该算法将三种特征进行融合,并采用Real_Adaboost分类器进行训练,以输出每个像素属于轮廓或者其他区域的概率。最后,采用条件随机场(CRF)对分类器输出结果进行后处理,以获得更加准确的篡改轮廓线。根据标准数据集CASIA2、Columbia上的测试结果显示,该算法的分类与定位性能均优于传统的基于手工设计特征的拼接检测算法。

Blind Forensics for Image Tampering Based on Tampered Contour

In order to solve the problems of traditional image mosaic tampering detection methods, such as single feature, solidification of detection methods and large positioning deviation of tampered areas, a tamper detection method based on tampered contour is proposed. In this method, image mosaic tampering traces are analyzed and modeled in frequency domain. Multi-step Markov feature is extracted to characterize the transfer characteristics of coefficients, and local information entropy and differential excitation are extracted to characterize the current distribution of position coefficients. The algorithm fuses the three features and trains them with Real_Adaboost classifier to output the probability that each pixel belongs to contour or other regions. Finally, conditional random field(CRF) is used to post-process the output of the classifier to obtain more accurate tampered contours. According to the test results on standard data sets CASIA2 and Columbia, the classification and location performance of the algorithm is better than the traditional detection algorithm based on manual design features.