| 摘要 |
Disclosed is a steganalysis method based on local learning. The method comprises: constructing a training sample database comprising positive samples and negative samples; for any to-be-detected sample, searching, in a sample database, for K positive and negative sample pairs most similar to the to-be-detected sample, so as to form a local training set; performing classifier training learning on the local training set, and adding the constraint of the positive and negative sample pairs in a learning process, so as to obtain an optimal classifier by using an optimization algorithm; and discriminating and classifying the to-be-detected samples by using the obtained the classifier, so as to obtain the detection result indicating whether steganography is performed on the to-be-detected samples. By making full use of local learning that has the advantages of better overcoming the problem of great intra-class change, reducing noise influence and requiring less priori knowledge, the present invention improves the effect of the steganalysis, and the present invention can be used in a mode recognition-based steganalysis algorithm analysis system. |
