基于边缘CFA内插特征一致性的图像拼接检测

Image splicing detection based on consistency of edge-based CFA interpolation features

为了有效检测数字图像是否被篡改,提出了一种新颖的数字图像拼接快速检测算法.首先,利用Canny边缘检测算子,依据阶跃性约束条件,筛选出阶跃型边缘像素点.分别针对自然图像中的阶跃型边缘和由拼接操作所形成的阶跃型边缘,建立垂直于边缘方向单像素序列的颜色滤波阵列(CFA)内插模型.通过计算该单像素序列的局部灰度直方图,设计CFA内插特征量,以测度拼接前后灰度水平级在直方图两端分布的差异.然后,基于此特征量,利用阈值法实现对自然边缘像素点和拼接边界像素点的分类,进而实现图像拼接检测与定位.标准拼接图像数据集上的测试结果表明,虚警概率仅为0.05时,本检测算法仍能达到0.97的高准确率.此外,该算法还具有较好的鲁棒性,可抵抗一定程度的JPEG压缩和模糊等后处理.

A novel digital image splicing detection algorithm is proposed to detect image tamper. First, step edge pixels are screened out by Canny a detector and step constraint conditions. A color filter array （CFA） interpolation model of a single-pixel sequence is created for the natural step edges and the splicing step edges. Such a sequence centers at an edge pixel and arranges along the vertical direction of the located edge. By calculating the sequence＇s local gray-level histogram, a CFA interpolation metric is designed to measure the distribution discrepancy between the natural edges and the splicing edges. Finally, a threshold-based method is employed to classify the natural edge pixels and the splicing edge pixels, and further to detect and discover the splicing trails. The results on a standard test dataset show that with a low false positive rate of 0. 05, the precision of the proposed method can reach as high as 0. 97. Besides, the splicing detector is so robust that it can resist post-processing such as JPEG compression and blur operation to a certain extent.