Computing and Informatics

Existing deep-learning-based image steganalysis networks have problems such as large model sizes, significant runtime memory usage, and extensive computational operations, which hinder their deployment in many practical applications. To address these challenges, we applied model compression techniques to image steganalysis and designed a model called SASRNet, a slimming-assisted steganalysis residual network. We observed that the trainable scale factor of BN (batch normalization) layer in steganalysis network can be used as channel scaling factor for pruning. The channel-level sparsity of convolutional layers is enhanced by imposing L1 regularization on channel scaling factors and pruning less informative feature channels. With the goal of balancing performance and efficiency, the iterative algorithm is used to further compress the network to obtain a slimming- steganalysis detector. In contrast to many existing methods, our proposed method can be directly applied to steganalysis network architectures by introducing a minimal overhead to the training process. We have conducted extensive experiments on BOSSBase+BOWS2 dataset. Experiments show that, compared to the original steganalysis model, this method can achieve comparable performance with less than 5 % of the parameters, validating the feasibility and practicality of the new model.