A Novel Hybrid framework of NAdamBound optimized Dilated Depthwise Separable CNN for deep learning based image steganalysis in digital forensics

Abstract

Introduction: Steganography is the method of hiding confidential information in different types of media, such audio, video, or photographs, while steganalysis is the process of finding and removing that information. Due to their capacity to automatically extract hierarchical features from input data, deep learning methods—in particular, those based on Convolutional Neural Networks (CNNs)—have recently surpassed conventional machine learning techniques in steganalysis tasks.

Objectives: CNN-based models frequently suffer from problems like overfitting, excessive power consumption, and expensive computational costs. These drawbacks make it difficult to use them in real-world situations, particularly when working with sizable or authentic datasets. In order to overcome these issues, this study attempts to create an accurate and effective CNN-based steganalysis model.

Methods: The suggested model DDS_SE-NB-Net incorporates dilated convolutions into a Depthwise Separable Convolutional Neural Network (DS-CNN) supplemented with Squeeze-and-Excitation (SE) blocks to capture multi-scale spatial data effectively while maintaining computational efficiency. To increase training stability, speed up convergence, and improve generalization, the model makes use of the NAdamBound optimizer, which combines the advantages of adaptive learning rate bounds with Nesterov momentum. In order to maintain high classification accuracy, the ideal dilation rate is also carefully chosen.

Results: The proposed DDS_SE-NB-Net achieved impressive accuracy rates of 94.0%, 92.2%, and 93.8% in detecting steganographic content embedded using the WOW, S-UNIWARD, and HILL algorithms, respectively. These results demonstrate a significant improvement in performance compared to existing CNN-based architectures, particularly on real-world stego image datasets.

Conclusions: A user-friendly Python-based Graphical User Interface (GUI) is built using Tkinter. The GUI enables users to upload images, initiate classification, and instantly view the steganalysis results, thereby showcasing the model's real-world applicability in digital forensics. This integrated solution illustrates the practical potential of the proposed system in aiding law enforcement and forensic investigators in detecting covert communications.