Automatic Recovery of Hidden Image from Image Steganography Using DNN and Local Entropy Features

Image steganography hides secret information in an image called cover image so naturally that the other users can not recognize the existence of information in the revealed image. This paper deals with an approach to recover the hidden image information from image steganography. The proposed approach investigates that the decoded hidden image information is a normal image or not. The normal and incorrectly decoded abnormal images have been trained using a deep neural network model and entropy features. The discrimination is processed with image patches since the information may be partially embedded in the cover image. The experiments are performed with respect to the various data capacities. The proposed approach discriminates and recovers the hidden image information automatically from a tremendously large number of steganography encoding methods.