Abstract: With the advent of image editing software，digital image can be easily manipulated and altered.If the image is tampered maliciously，the consequence may be very serious.So the detection of image tampering has become more and more important.Most digital cameras employ a single sensor and a color filter array，and then interpolate the missing color samples to obtain a color image.The interpolation introduces specific correlations and the correlations may be destroyed by the image tampering.EM algorithm can be used to analyze the correlations.Through the improvement of the existing model，the Gaussian distribution of the residual error is supplanted by the Laplace distribution in order to approach the real distribution.The results of experiment show that this approach can reveal traces of digital tampering in color images interpolated with different algorithms and the improved model can speed up convergence of EM algorithm and reduce the detecting time.

摘要： 随着图像编辑软件的普及，数字图像可以很容易地被编辑和修改，但如果恶意篡改，后果可能会非常严重，因此图像篡改检测越来越重要。大部分数码相机都使用单传感器和颜色滤波阵列，并通过插值获得彩色图像。插值引入了相关性，而篡改操作会破坏这种相关性，可以使用EM算法对其进行检测。通过对已有模型进行改进，用拉普拉斯分布代替插值点残留误差的正态分布，使分布情况更符合实际。实验证明该算法可以检测出不同插值算法下的篡改痕迹，并且改进后的模型加快了EM算法的收敛速度，缩短了篡改检测的时间。