【摘要】 针对数字图像取证技术中的相机来源鉴别问题,利用相机成像过程中颜色滤波阵列(CFA)插值的线性模型,采用协方差矩阵对CFA插值系数进行统计估计,有效地降低了现有方法存在的估计误差,并以估计的CFA系数为特征,利用SFFS方法从240维特征中优化选择了36维组成特征向量,最后使用支持向量机作为分类器,对22种不同品牌不同型号的相机拍摄的图像进行了来源鉴别。实验表明,本文算法对数字图像的相机来源鉴别检测,其平均正确率达到了96.5%,优于现有的相机来源鉴别算法。更多还原

【Abstract】 Focusing on the problem of source camera identification,which is an important branch of digital forensices,a novel approach is proposed in this paper.By solving the covariance matrix equation,a statistical estimation of CFA interpolation coefficients is applied to reducethe bias.The feature selection process is implemented with SFFS method to construct a vector of 36 features selected from 240 coefficients,and the vector is then fed to the SVM classifier.The experiments,including 22 cameras consist of different makes and models,show a higher accuracy of 96.5% of our method on source camera identification,compared with the existing ones.更多还原