Performance evaluation of feature-based steganalysis in steganography

The objective of this paper is performance testing of Feature-based Steganalysis methods for detection of steganography tools that are used for hiding a secret message in still images. Many of those tools, which can be used publicly, cause statistical changes in original images during embedding process. The features represent those statistically calculated changes, where feature extraction in this paper was applied in spatial domain and also directly in transformation domain of DCT in JPEG files what helps to obtained relevant statistical data. The results of this paper identify the selection of statistical feature vector that is used during training phase of classifier in order to distinguish between cover and stego image. The reliable detection of selected steganography methods were presented in relation to length of feature vector. Moreover, contribution to design of blind steganalysis system was proposed for the purpose of unveiling a secret communication via completely new steganography methods.