Length Estimation for Embedded Message Using Support Vector Machine

Regular singular (RS) and sample pair analysis (SPA) steganalysis methods are highly sensitive to message length which is embedded in stego-image. Nevertheless, they are only applied to spatial domain. Lyu and Farid enhance universal steganalysis methods to detect the stego-image both in spatial and frequency domains, but it could not estimate the length of embedded message. A novel steganalysis method possesses the function of message estimation is presented by combining multi-classification with support vector machines (SVMs). It could also be applied to both spatial and frequency domains. Experimental results show that the efficiency and practicability of the proposed scheme is more feasible than RS and SPA steganalysis methods.