Image steganalysis based on model compression

Abstract

Deep learning technology has developed rapidly in recent years, and deep learning-based steganography and steganalysis techniques have also achieved fruitful results. In the past few years, the over-expanded structure of steganalyzers based on deep learning has led to huge computational and storage costs. In this article, we propose image steganalysis based on model compression, and apply the model compression method to image steganalysis to reduce the network infrastructure of the existing large-scale over-parameter steganalyzer based on deep learning. We conducted extensive experiments on the BOSSBase+BOWS2 dataset. As can be seen from the experiment, compared with the original steganalysis model, the model structure we proposed can achieve performance with fewer parameters and floating-point operations. This model has better portability and scalability.