Analysis of Deep Learning-Based Image Steganalysis Methods Under Different Steganographic Algorithms

In this paper, we present an experiment to evaluate the performance of deep-learning-based steganalysis methods under different steganographic algorithms and payloads. We designed and implemented a simple and efficient convolutional neural network (CNN) steganalysis model. This model combines SRM [10] filters and 2D Gabor filters to initialize the CNN's filters in the feature extraction layers. The extracted features are passed to a fully connected neural network classifier to detect images containing hidden messages. We generated several datasets using different steganographic algorithms and payloads to train and test our CNN model. We designed several experiments to evaluate the detection rate of the proposed model under separate training-testing conditions.