Steganalysis of an Image-Based Semantic Segmented and Binary Pattern Complex Stego File Using RNN And DenseNet

Authors

  • J. N. Ugwu, T. T. Odufuwa, S. E. Akinsanya and K. V. Fajembola Department of Computer Science, Federal University Oye-Ekiti, Nigeria

Keywords:

Deep Learning, Image Steganalysis, CNN, LSTM, Steganography

Abstract

Image steganalysis is a method for detecting concealed data within an input image, regardless of the steganography technique employed. Unfortunately, many existing image steganalysis techniques need more detection accuracy and high computing costs due to preprocessing requirements. This paper proposes a hybridized CNN-RNN deep learning approach that leverages DenseNet and LSTM techniques for efficient steganalysis. Using a combined embedded and unaltered image dataset of about three hundred thousand which are partitioned into training and testing sets with the ratio of 70 to 30 percent.  The method utilizes the dense concatenation property of DenseNet and the high sensitivity of LSTM to identify differences in image appearance and extract concealed data from the cover image. Performance analysis was conducted on both models using mean squared error (MSE), peak signal-to-noise ratio (PSNR) and accuracy. The results demonstrate that LSTM outperforms DenseNet regarding MSE and PSNR values, LSTM shows lower MSE and higher PSNR The ensemble of both model gives higher accuracy values indicating a more quality image after extraction from an embedded form, thus making it a superior model for image steganalysis. This approach offers an effective and efficient method for detecting concealed data in image steganography. LSTM provides a promising avenue for further improvement of image steganalysis techniques. The findings will contribute to the advancement of steganalysis research and pave the way for practical applications in the field of information security.

Author Biography

J. N. Ugwu, T. T. Odufuwa, S. E. Akinsanya and K. V. Fajembola, Department of Computer Science, Federal University Oye-Ekiti, Nigeria

Downloads

Published

04/30/2024