Blind Steganalysis using One-Class Classification

Steganography is the science/art of hiding information in a way that must not draw attention to the message hidden in the transmitted media, if a suspicion is raised then there is no meaning to the purpose of steganography. Then appeared its counterpart, Steganalysis, which aims to suspect and analyze the transmitted media to decide wither it contain an embedded data or not which we present in a blind Steganalysis way. One-Class Classification (OCC) machine learning algorithms aim to build classification models depending on positive class only when the negative class is not available or poorly sampled. Here in this paper we depend on a one-class support vector machines (OCSVM) which has been trained on only one class of images that is clean images class, so that the trained classifier can classify new reviews to their correct class i.e. clean or stego. Training an OCC turned to be hard work and required long execution time since classifier parameters tuning, data separation and model evaluation needed to be done manually in a brute force way. A powerful programming language (Python) with the powerful machine learning library (Scikit-Learn) gave a promising classification results in deciding whether an input image is clean or stego image.