基于图像背景噪声特性的篡改检测

数字图像都包含有一部分来自成像过程或者数字压缩的背景噪声，如果两幅不同背景噪声的图像被拼接在一起，则图像篡改区域和其他区域的噪声特性会有差异。本文基于一种估计信道信噪比的高阶统计量法提出了一种新的图像背景噪声的盲估计算法。通过对图像进行分块计算每块的噪声方差，从而检测图像篡改部分。此算法通过二次加噪的方法解决了高阶统计量法中必须已知原始信号的问题，实现了待检测图像噪声的盲估计。实验结果显示该算法能有效估计图像的噪声方差从而达到检测局部篡改的目的。并且图像的缩放和压缩对检测结果影响很小，算法具有较好的鲁棒性。 

Image forgery detection using characteristics of background noise

There is a part of background noise in digital image which comes from imaging process or digital compression. The noise characteristics between image forgery area and the other area would be different if images with different noise levels are spliced together. Base on a method of channel SNR estimation which uses high order statistic characteristics, a new algorithm is proposed in this paper for blind estimation of image background noise. We can detect the forgery parts by dividing image into some sub-blocks and computing the noise variance of each ones. The problem of unknown original signal in the method of high order statistic characteristics has been solved by adding noise again and the blind estimation of detected image background noise has realized. Experiment results show the algorithm is effective in estimating noise variance and detecting forgery parts. In addition, there is little influence on the experiment results after image scaling and compression. So the algorithm has good robustness.