中文摘要：

针对由模糊图像拼接而成的合成图像，提出了一种基于盲源分离的被动取证新方法．该方法基于模糊合成图像的倒谱特性估计出模糊参数，根据重构的模糊函数对其进行盲复原，通过对复原产生的振铃效应进行测量实现了混合矩阵的估计，从而完成了篡改区域和未篡改区域的分离．实验结果表明，该方法针对经历不同拼接操作的模糊合成图像均取得较好的检测效果，同时，与现有的方法相比，本文算法对高斯白噪声和有损JPEG压缩具有更好的鲁棒性．

英文摘要：

This paper proposes a passive forensics method for spliced blurred image based on blind source separation （ BSS ）. The blur parameters are estimated by the characteristics of spliced image in the cepstrum domain, and then the blurring kernels are constructed and the spliced blurred image is restored. This paper also develop a new measure scheme of ringing effect in order to estimate the mixing matrix and then recover all source images, making us able to identify the spliced region and the background region. Experi-mental results show a high detection accuracy on the spliced image with various blurting operations. Compared with other existing al-gorithms, the proposed method has better robustness against gaussian noise and lossy JPEG compression.