为了实现无先验信息情况下的CT图像来源检测,提出了一种基于CT扫描仪原始传感器模式噪声的来源检测算法.首先,采用由5种不同滤波器构建出的滤波器组,从CT图像中获取5种不同类型的传感器模式噪声;然后,对每一种噪声分别进行CT三维重建反变换,得到原始传感器模式噪声;最后,根据原始传感器模式噪声的统计特征向量,并结合支持向量机分类器进行分类,从而实现来源检测.实验结果表明,针对4个厂家的15种不同型号CT扫描仪,采用所提算法获得了较基于模式噪声的来源检测算法更高的识别精度,其平均分类精度可达94.12%.
To detect the origin of a computed tomography (CT) image without prior knowledge, an origin identification algorithm based on the original sensor pattern noise of the CT scanner is proposed. First, 5 different sensor pattern noises are extracted from CT images by a filterbank constituted of 5 distinct filters. Then, the original sensor pattern noises are obtained by the inverse transform of three dimensional CT reconstruction for each type of noise. Finally, the statistic feature vectors of the original sensor pattern noises and the support vector machine (SVM) based classifier are combined to tackle origin identification. The experimental results show that as for 15 different CT scanners from 4 manufacturers, the proposed origin identification algorithm obtains higher identification precision compared with other existing origin identification algorithms based on pattern noises, and the average classification accuracy can reach 94. 12%.