Enhancing Cybersecurity: Detecting Hidden Information in Spatial Domain Images Using Convolutional Neural Networks

Akram Mshet^1,* , Huda Tayyeh²

¹Informatics Institute for Postgraduate Studies Iraqi Commission for Computers & Informatics, Baghdad, Iraq

²University of Information Technology  and Communications, Baghdad, Iraq

Emails: 2020dhiqar@gmail.com; haljobori@uoitc.edu.iq

Abstract

Steganography involves concealing hidden messages inside various types of media, whereas steganalysis is the process of identifying the presence of steganography. Convolutional neural networks (CNN), a type of neural network that outperformed previously proposed machine learning-based methods when introduced, are among the models used for deep learning. While CNN-based methods may yield satisfactory results, they face challenges in terms of classification accuracy and network training stability. The present research introduces a CNN structure to increase hidden data detection and spatial domain image training reliability. The suggested method includes pre-processing, feature extraction, and classification. Evaluation of performance is conducted on datasets Break Our Steganographic System Base (BOSSbase-.01) and Break Our Watermarking System (BOWS2) with three adaptive steganography algorithms. Wavelet Obtained Weights (WOW), Spatial Universal Wavelet Relative Distortion (S-UNIWARD), and Highly Undetectable steGO (HUGO) operating at low payload capacities of 0.2 and 0.4 bits per pixel (bpp). The experimental results surpass the accuracy and network stability of prior publications. Training accuracy ranges from 91% to 94%, and testing accuracy ranges from 74.8% to 86.65%.

Keywords: Deep learning; Steganography; Convolutional neural network; Steganalysis