A JPEG steganalytic method based on semi-supervised learning algorithm was presented.Using three catego-ries of statistical features for JPEG images and multiple hyperspheres one-class SVM

three classifiers were generated from the original labeled example set.These classifiers were then refined using unlabeled examples in the Tri-training process

which could effectively improve detecting ability by exploiting a large amount of unlabeled images.Experimen-tal results showed the effectiveness of our proposed method.