Impact of Noise and Complexity on Targeted Image Steganalysis

Abstract

The changes made by steganography to an image can be considered as noise addition. A fundamental question is how well targeted steganalytic methods distinguish between the stego signal and naturally occurring noise in images. Moreover, several researchers have proposed steganographic methods that hide the message data in the complex regions of images. They argue that those regions are more secure for data hiding. However, no systematic experiments, to the best of our knowledge, have been presented to confirm the claims. This paper provides an experimental study to answer the questions about the impact of noise and complexity on the performance of targeted steganalytic methods. Three well-known targeted steganalytic methods are tested when spatial domain steganography is considered. Three types of noise at different levels are added to images, and two groups of complex and non-complex images are considered. Our experiments show a significant decrease in the detection accuracy of all the selected steganalytic methods for complex images. The methods demonstrate different responses to different noise types. Their performance degrades the most by Gaussian, and the least by salt & pepper noise.