| Image steganalysis is an art of detecting hiding information in digital image. Recently, steganalysis has been paid a great deal of attention both from the academic research and law enforcement. On one hand, steganalysis can improve security of steganographic algorithm. On the other hand, it can be used to reveal the illegal communication with some steganographic tools. The universal steganalysis for images with passive adversaries are focused on in this thesis. The main contributions are summarized as follows.1. A universal steganalysis method for the additive noise modelable steganography in the special domain is put forward. Higher order statistics moments of the differential histogram distributions and the ratios of special areas under the differential histogram curves are calculated to characterize the difference between the differential distribution of the cover image and that of the stego-image. Fisher classifier is used to detect whether there exists the hidden information in images. Experimental results show that our technique achieves better detection performance.2. A universal steganalysis method is designed for the steganography techniques in the JPEG compression domain. The features for the blind classifier are calculated as the mean, variance, skewness, kurtosis of the histogram distribution of the DCT coefficients, and the histogram distribution probabilities of special coefficient values as well. The variance analysis is utilized to identify useful features in steganalysis. Experimental results prove the validity of the method. |
