JPEG Steganalysis Based on ISODATA and SVM
Tianjie Zhu, L. N. Wang, Yizhu Ren, Mengying Wang · International Conference on Multimedia Information Networking and Security · 2013
A novel JPEG universal steganalysis algorithm, based on ISODATA and SVM, is proposed in this paper. ISODATA clustering algorithm is employed to divide the trained images into some clusters, before the images are used to train supervised machine learning algorithms. The images of each cluster are trained by using SVM subsequently and detection model is constructed individually. The minimal distance from test images to center of cluster is used as the criterion to choose the model. The experiment shows that the detection accuracy of method based on ISODATA and SVM is better than the method used SVM only and the accuracy of steganalysis algorithm can be improved by using the ISODATA to divide the training examples sets beforehand.