Geometrical Self-Organizing Map Classifier Based on Active Learning for Steganalysis in the Video Environment by Spending at Least a Label

Classifier is one of the three blocks of a video steganalysis that needs labeled for training. In the blind video steganalysis, due to the lack of access to steganography algorithms, it is difficult to label. In this paper, the semi supervised growing self-organizing map classifier has been used to reach the minimum label. For this purpose, a concept called the geometric redundancy of the lower-layer nodes of the semi supervised self-organizing network has been used. It has been shown that this redundancy will create repetitive patterns of the network, so deleting such nodes is possible. Proven due to the existence of one-to-one correspondence between nodes and labels. Reducing nodes leads to a reduction in the number of labels required. The basic point is the need for a geometric redundancy among a number of nodes, which is a conception of abstraction, is the formation of a group by them. Therefore, the proposed algorithm is based on identifying categories and integrating their members. The classifier obtained on this basis has been named a geometric self-organizing map classifier .It is proven that this classifier can achieve the minimum amount of optimal label. The simulation results show a remarkable superiority over the previous algorithms.