LAUSR

There are few studies on binary image steganalysis based on convolutional neural network (CNN). In this paper, an efficient binary image steganalysis scheme based on CNN which integrates high-pass filters, truncated linear unit and subnetworks is proposed. In the process of binary image steganography, flipped pixels usually scatter on the boundaries of the content in the image. Therefore, the first convolutional layer is constructed with high-pass filters to capture the structure of embedded signals better. Truncated linear unit (TLU) is also adopted after the first convolutional layer for the same purpose. 4 truncated linear units with different truncated values are adopted to capture embedding signals of different intensities. We also adopt 4 subnets after the 4 truncated linear units to further boost the performance of the CNN network. The experimental results show that our proposed scheme is efficient and effective on binary steganalysis.