Steganography detection using RBFNN

A machine learning approach based on alpha-trimmed mean feature preprocessing is introduced to determine whether secret messages are hidden within JPEG images. This paper also integrates a multi-preprocessing sequence to develop the classification system which contains features generated from an image dataset including steganographic and clean images, feature ranking and selection, feature extraction, and data standardization. Neural networks using radial basis functions train the classifier to accomplish the decision making progress. The analyzed image is labeled as either a steganographic or a clean image. The computer simulations have shown that classification accuracy increases by 40% when using feature preprocessing within the complete detection system over a system without feature preprocessing. In addition, alpha-trimmed mean (including mean and median) statistics approach results in higher classification accuracy.