基于QDCT马尔科夫方法的彩色图像拼接检测

Splicing Detection for Color Images Based on QDCT Markov method

提出一种基于四元数离散余弦变换(quaternion discrete cosine transform,QDCT)的拼接检测方案.该方案充分利用了彩色图像三通道之间的相关性,减少了图像彩色信息的损失.将彩色图像进行QDCT得到变换系数,然后使用四元数马尔科夫方法提取特征,最后用支持向量机检测拼接图像.使用图像库CASIA TIDE v1.0和CASIA TIDE v2.0进行实验,所得准确率分别达到98.75%和96.78%,表明该方法优于大部分现有方法.

We propose a splicing detection scheme based on quaternion discrete cosine transform （QDCT）. The scheme uses correlation among three channels of a color image to reduce loss of the image＇s inherent color information. QDCT is first applied to the images, and features are extracted with the proposed scheme in the QDCT domain. SVM is used to detect image splicing. Using the image bases CASIA1 and CASIA2, accuracy of the proposed scheme reaches 98.75% and 96.78% respectively, which is better than most of existing methods.