A Modified Blind Steganalysis Method Based on the Moments of Characteristic Function

Abstract

Steganography is the way of protecting secret information which satisfies the needs of individuals and public security communication. It has become a powerful tool for criminals engaged in crime activities. Consequently, steganalysis is required for preventing these crime activities and also to measure the performance level of the steganography tools. In this paper, a modified blind steganalysis method is proposed for detecting image steganography through the prediction error image generation, transformation, feature extraction and classification processes. During the generation of prediction error image of the cover or stego image, a modified prediction algorithm is used to get better detection rate. Then, three order haar wavelet transform is performed on the image and its prediction error image to produce twelve wavelet subbands. Moreover, the decomposition of first scale diagonal subband is also applied to get four extra subbands. In feature extraction process, the first three order moments of characteristic function of wavelet subbands (including image itself and its prediction error image) are selected to form the feature vectors. Finally, the classification task is performed to classify the image into either cover or stego image using feed-forward Backpropagation artificial neural network (ANN). The presented method is evaluated for various steganography tools such as StegJ, Openstego, Image Steganography and invisible Secret. The comparative experimental results show that the classification rate of the proposed steganalyzer is 60-90% and the improvement is up-to 20%.