摘要
重采样是图像篡改操作的方法之一.JPEG图像经过重采样之后,DCT系数与邻近的系数之间会发生变化,因此利用多变量广义高斯分布模型来逼近DCT系数的概率分布,根据不同重采样图像的DCT系数邻近联合概率密度分布的不同从而将其作为重采样检测的首选特征,然而仅采用DCT系数邻近联合概率特征易受到图像纹理的影响从而降低重采样的检测率.而Pseudo-Polar分数傅立叶变换是将DCT系数所在的笛卡尔坐标系映射到pseudo-polar坐标,不同的重采样操作对其进行分数傅里叶变换会产生不同的纹理,所以提取其光滑度作为DCT系数特征的补充.最后,利用SVM对两类特征进行训练以及检测,实验结果表明,对于缩放因子大于1以及旋转角度较大的重采样图像有较高的检测率以及较强的鲁棒性.
Resampling is one of the important methods about image tamper operations. The linear relationship between coefficients and joint DCT coefficients is affected when the image is compressed after resampling,thence the multivariate generalized Gaussian distri- bution model is used to approximate probability distribution of DCT coefficients, so neighboring joint density of DCT coefficients is different with different resampling operations and is made as the preferred characteristic,but resampling detection rate is affected by the image texture when only neighboring joint density is used. Pseudo-Polar Fractional Fourier Transform can make Cartesian coordinate map to the pseudo-polar coordinates ,however ,different textures is generated when fractional Fourier transform is applied to DCT coef- ficients, so the smoothness is extracted as supplementary for DCT coefficient characteristics. At last, SVM is used to train and test the two types of features, the experiment shows that the algorithm proposed in this paper has a good detection for resampling images whose scaling factor is greater than 1 and larger rotation angle,it also has strong robustness.
出处
《小型微型计算机系统》
CSCD
北大核心
2015年第4期868-871,共4页
Journal of Chinese Computer Systems
基金
国家自然科学基金项目(61373126)资助
