摘要： 现有文献对具有旋转等几何变换的复制粘贴篡改操作检测能力有限。为此，提出基于径向矩参数估计与奇异值分解的图像复制粘贴快速取证算法。通过构造本身具有旋转不变性的圆形结构，采用高斯金字塔分解用以降低图像尺寸，利用图像块奇异值分解构造七维特征向量，块匹配字典排序法以减少特征向量的搜索匹配空间，并采用径向矩参数估计旋转参数，用主旋转角度计替代主向量转移方法去除误匹配块。实验结果证明，圆形结构可解决图像因旋转而造成的像素错位现象，基于径向矩的旋转参数估计能提高算法的检测精度，减少误匹配块。而由奇异值组成的特征向量对图像的后处理操作JPEG压缩、加性噪声等具有较强的鲁棒性。

关键词: 图像盲取证, 旋转不变性, 奇异值分解, 径向矩, 复制粘贴伪造, 取证检测

Abstract: Existing algorithms have not enough ability to detect the copy move forgery. In order to improve the accurate of the current algorithms, the new detection is proposed by constructing the circles replacing the traditional ways which are based on the square. The seven characterizes are constructed according to singular value decomposition. The dictionary-ordering method is applied for saving the time-consuming. Using main rotation angle based on the radial moment and the proportion of constraint remove the error mark. Experimental result shows that circle structure can solve pixels mismatching perfectly. The radial moment is applied to improve the accuracy of detection. Meanwhile seven characterizes are robust for post-processing, such as JPEG compression and additive noise.

Key words: blind image evidence obtain, rotation invariant, singular value decomposition, radial moment, copy move forgery, evidence obtain detection

中图分类号: 

• TP391.41