HSDetect-Net: A Fuzzy-Based Deep Learning Steganalysis Framework to Detect Possible Hidden Data in Digital Images

Abstract

Recent progress in digital image steganalysis has been heavily influenced by advancements in Deep Learning (DL), particularly using Convolutional Neural Networks (CNNs). These networks have become a preferred choice for addressing complex classification challenges. Despite their success, current CNN-based approaches for steganalysis often encounter difficulties in accurately identifying concealed information, especially when dealing with complex textures generated by steganographic techniques. To overcome these issues, this study introduces HSDetect-Net, a novel CNN framework that integrates fuzzy logic to enhance detection capabilities. The proposed HSDetect-Net's architecture employs specialized small-sized convolutional kernels to extract complex details and incorporates a fuzzy layer to refine classification accuracy. Evaluations conducted on the Break Our Steganographic System Base (BOSSBase) and Break Our Watermarking System (BOWS) datasets reveal that HSDetect-Net achieves remarkable performance, with an accuracy of 99.07% and an F1 score of 99.68%. Compared to existing methods, HSDetect-Net improves the accuracy of detecting hidden data in images, making it a promising solution to the steganalysis of digital images.