高级检索
    王 波 孔祥维 付海燕. 联合OC-SVM和MC-SVM的图像来源取证方法[J]. 计算机研究与发展, 2009, 46(9): 1456-1461.
    引用本文: 王 波 孔祥维 付海燕. 联合OC-SVM和MC-SVM的图像来源取证方法[J]. 计算机研究与发展, 2009, 46(9): 1456-1461.
    Wang Bo, Kong Xiangwei, and Fu Haiyan. A Source Camera Identification Method Based on the Combination of OC-SVM and MC-SVM[J]. Journal of Computer Research and Development, 2009, 46(9): 1456-1461.
    Citation: Wang Bo, Kong Xiangwei, and Fu Haiyan. A Source Camera Identification Method Based on the Combination of OC-SVM and MC-SVM[J]. Journal of Computer Research and Development, 2009, 46(9): 1456-1461.

    联合OC-SVM和MC-SVM的图像来源取证方法

    A Source Camera Identification Method Based on the Combination of OC-SVM and MC-SVM

    • 摘要: 为了解决现有图像来源取证方法在相机样本较多时准确性较差、无法对未知模型的图像来源取证以及可扩展性差的问题,提出了一种基于一类和多类支持向量机联合的图像来源取证方法.算法利用协方差的统计相关性提高了CFA插值系数的估计精度,并以SFFS算法选择的特征作为分类器输入.采用OC-SVM(一类支持向量机)和MC-SVM(多类支持向量机)联合的策略进行图像来源分类,有效地解决了对未知模型图像来源的鉴别问题以及可扩展性差的问题.实验表明,该方法对28种相机拍摄的图像进行来源取证,能够达到平均90.4%的鉴别正确率,同时对于3种训练模型以外的未知相机模型拍摄图像,能够达到平均79.3%的检测正确率.

       

      Abstract: The multi-class classifier used in the existing source camera identification algorithms usually leads to numbers of problems, such as unavoidable false classification of the images out of the training models, decreasing accuracy as camera models increasing and the lack of expansibility. Focusing on these problems, a method for source camera identification based on the combination of one-class SVM and multi-class SVM is proposed in this paper. By solving covariance matrix equation, the authors reduce the perturbing term introduced by the pipeline of imaging, and improve the estimating precision of CFA interpolation coefficients. To obtain a more efficient feature space for classification, the sequential forward feature selection method is implemented to construct feature vector as the input of the classifier. The strategy using the combination of OC-SVM and MC-SVM as the classifier in the approach provide an effective approach for the classification of images out of training models and system’s expansibility. In the combination, the OC-SVM is used to expose the images that captured by an unknown camera model, and the MC-SVM trains a new multi-class model to classify the image source according to the positive results of the OC-SVM. The experiments indicate that average accuracies of 90.4% for camera model identification from 28 cameras, and 79.3% for three outlier camera model detection are obtained respectively in this method.

       

    /

    返回文章
    返回