Detection of Malicious Spatial-Domain Steganography Over Noisy Channels
Steganographic channels can be abused for malicious purposes, thus raising the need to detect malicious embedded steganographic information (steganalysis). This chapter will cover the little-studied problem of steganography and steganalysis over a noisy channel, providing a detailed modeling for the special case of spatial-domain image steganography. It will approach these issues from both a theoretical and a practical point of view. After a description of spatial-domain image steganography, the impact of Gaussian noise and packet loss on the steganographic channel will be discussed. Characterization of the substitution-insertion-deletion (SID) channel parameters will be performed through experiments on a large number of images from the ALASKA database. Finally, a steganalysis technique for error-affected spatial-domain image steganography using a convolutional neural network (CNN) will be introduced, studying the relationship between different types and levels of distortions and the accuracy of malicious image detection.