Detection of stego images using one-class support vector machines with multiple hyperspheres
多超球面OC-SVM算法在隐秘图像检测中的应用

In order to reduce the complexity and weak generalization of classification method using two class support vector machines in images steganalysis, a new Kernel-based classification method using OC-SVMs with multiple hyperspheres was put forward. Considering that the data features were expected to be more separable in kernel space, we first performed the K-means clustering in kernel space, then trained the sub-class data separately using OC-SVMs and established a multiple hyperspheres classification model to decide the class label of new data. The experimental results show that this method has efficiently improved the classification precision.