SNENet: An adaptive stego noise extraction network using parallel dilated convolution for JPEG image steganalysis

The steganalysis for JPEG image is an important research topic, as the enormous popularity of on Internet. However, stego noise feature extraction process existing deep learning-based steganalytic methods are not adaptive enough to content image, which may lead suboptimal performance. In order solve this issue, network, named SNENet, proposed. module network specifically designed steganalysis, consists parallel dilated convolutional layer and inverted bottleneck layer. This specific design expands receptive field makes more global image. experimental results indicate that proposed outperforms state-of-the-art method by much 6.25% UED-JC 3.35% J-UNIWARD. also justified in extensive ablation experiments.