鲁棒的复制-缩放-粘贴篡改检测方法

Robust copy-scale-move forgery detection method

为了能够检测图像的复制-缩放-粘贴(CSM)篡改操作,提出了一种基于金字塔模型的CSM图像篡改检测方法.首先,将一幅测试图像缩放成不同尺度的降采样子图像,采用非均匀采样选择子图像,并建立金字塔篡改伪造检测模型;然后,对模型中每一层子图像进行非重叠分块,提取该模型所有图像子块的Zernike矩特征;最后,模型中每一层的所有块特征分别与底层特征建立KD树,以便于搜索图像块的所有近似匹配对,并从匹配对中寻找CSM篡改操作的伪造痕迹平移向量,以便于确定篡改区域.实验结果表明,与现有的图像篡改检测方法相比,所提方法获得了更好的检测结果,且篡改图像检测的查准率和查全率均超过93%.

In order to detect the image copy-scale-move （CSM）forgery,a CSM forgery detection method based on a pyramid model is proposed.First,a test image is rescaled into down sampling sub-images with different scale factors,which are used to build the pyramid model for forgery detec-tion according to non-uniform sampling.Then,each sub-image in the pyramid model is divided into non-overlapping sub-blocks.The Zernike moment features of all sub-block are extracted.Finally, the block features of each sub-image and those of the original image are used to build a KD-tree to search all approximate matching pairs,from which the forged trace shift vector can be found so as to determine the tampering region.The experimental results show that compared with the existing image tampering detection methods,the proposed method provides better detection results for large scaling attacks of region-duplication.The precision and recall of detection are more than 93%.