基于分形和统计的复制-粘贴篡改图像的检测

现在大多数图像“复制-粘贴”篡改检测算法对于区域复制后的进一步混合处理不能进行有效检测。为此提出了一种新的基于分形和统计的检测方法。首先将图像分块并提取每块的特征向量，该特征向量由分形维数和三个统计量组成；接着对所有特征向量进行字典排序；最后,利用图像块的位置信息和欧氏距离定位篡改区域。此方法不仅能够检测传统的复制-粘贴型篡改，而且还能够检测经过旋转、翻转以及旋转和翻转混合的多区域复制-粘贴型篡改；此方法也能够抵抗高斯模糊、对比度调整和亮度调整等攻击。实验结果表明了该方法的有效性。

Detection of copy-move forgery image based on fractal and statistics

Most of the existing detection algorithms for image copy-move forgery cannot effectively detect the sequential mixed image forgeries with regional duplication, so a new detection method based on fractal and statistics was proposed. The presented method first divided an image into overlapping blocks, and the each block would respectively be extracted to an eigenvector, which was composed by fractal dimension and three statistical data. Then, all the eigenvectors were lexicographically sorted. Finally, the forgery part was localized by means of the location information of the blocks and the Euclidean distance. The proposed method can not only detect the traditional copy-move forgery, but also detect the multi-region forgery for images subjected to rotation, flipping, and a mixture of these processing operations. The method is also robust to tampered images undergoing some attacks like Gaussian blurring, contrast adjustment, brightness adjustment, etc. The experimental results show the validity of the method.