Abstract: A new technology of passive authentication in image forgery is proposed. It obtains two kinds of sub-block collections by segmenting the test image with two methods. Wavelet transform is performed on each block of the collections. The approximate coefficients in the highest level of the wavelet transform which represent the sub-blocks are expressed as row vectors. The vectors of each collection form a matrix. The PCA method is used to compress the data of the two matrices respectively. Tamper detection is carried out on the compressed data（the principal components of PCA）. The tamper areas are marked with labels. The experimental results indicate that the proposed algorithm can not only decrease computational complexity, but also is of good robustness to blurring, JPEG operation and noise contaminating as well as the mixture of these operations. It can detect most tampered areas in copy-move tampered images which are badly contaminated.

Key words: wavelet transform, principal component analysis, image forgery, passive-blind detection

摘要： 提出了一种新的图像盲检测技术，该技术先对图像进行两次分块得到两个子块集，分别对这两个子块集中的子块进行小波变换，将最大变换尺度的小波近似系数以向量形式表示各子块，一个子块集组成一个矩阵，利用主成分分析方法（PCA）对这两个特征矩阵进行二次特征提取，利用Pearson相关系数法对二次提取后的子块特征进行篡改检测，标记出篡改块。实验结果表明，该技术在降低运算复杂度的基础上，不仅能较好地检测进行了多处复制粘贴篡改的图像，且在抗高斯模糊、JPEG有损压缩和噪声方面都有较强的鲁棒性，尤其在篡改图像经过滤波和加性噪声混合严重干扰后，仍能检测出大部分篡改区域。

关键词: 小波变换, 主成分分析, 图像篡改, 盲检测