Image steganalysis using artificial neural networks

Abstract

This research work presents the steganalysis of steganographic images under predefined way of hiding information. Hiding information in the cover image is done by different methods. However, steganalysis algorithms proposed in this research work are unaware of the methods in which procedure the information was hidden in the cover image. The proposed algorithms are based on Artificial Neural Networks (ANN) for finding out the presence of any hidden information. Steganography is the process of embedding text information or imageinformation into cover image. The embedding is otherwise called hiding information in a cover image. The cover image appears with no change in the content when looked at the embedded image. At the same time, the presence of hidden information can be identified and if possible, the original hidden information can be recovered for better interpretation. This research identified group of cover images and information images that analyses and modifies existing algorithms and proposed their implementation for steganalysis. There are situations in which the presence of hidden information can be identified but cannot be reconstructed as interpretable information. Existing literature has used mostly statistical methods to identify and reconstruct the original information. Very few works have been carried out using ANN algorithms. The scope of this research work considers existing ANN algorithms like 1. Back propagation algorithm (BPA) 2. Functional update back propagation algorithm (FUBPA) 3. Radial basis function (RBF) Supervised BPA has been considered and learning of the generated data has been done. Training of the network with different learning factors has been tried and finally a value for learning factor with value 1 has been selected. This indicates a standard convergence. As a development of the vii BPA algorithm, conditional BPA called functional update back propagation ealgorithm (FUBPA) have been developed and implemented.