Estimating cover image for universal payload region detection in stego images

With the increase in digital communication, images have become indispensable media for carrying obscured communication these days. Unraveling the presence or details of obscured transmission also known as steganography is termed as steganalysis. Passive steganalysis focuses only on classification of object as clean or adulterated. Finding the length of payload and regions where the message is hidden, in a stego object is a challenging task and forms a part of Active or Forensic Steganalysis. Many research-ers have worked towards finding the length of message, but not much work has been done on finding region where message is hidden. This motivated us to work towards finding the region of payload in stego images. Hence this paper proposes a novel technique for detection of payload regions in a stego image combining Denoising Autoencoder with Local Binary Pattern (LBP) Operator. Denoising Autoencoder is used to learn inter relationships between pixels of an image for estimating cover image from the given stego image. LBP operator capable of capturing local inter-relationships within neighborhood coordi-nates, have been used to compare corresponding blocks in input stego image and estimated cover image for unraveling payload locations. The major advantage of the proposed technique is that it is universal and it does not require cover image or knowledge about steganographic algorithm used. Accuracy of 96.5% and 97.7% respectively was achieved in the experimental study conducted with spatial and JPEG domain images from BOSSBase dataset.(c) 2022 The Authors. Published by Elsevier B.V. on behalf of King Saud University. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

Automated summary

This paper proposes a novel technique that combines denoising autoencoder with local binary pattern (LBP) operator to detect payload regions in stego images. The technique does not require a cover image or knowledge about the steganographic algorithm used. Experimental study achieved accuracies of 96.5% and 97.7% in spatial and JPEG domain images, respectively.