Image Steganalysis Scheme based on Fusion of CNN and BiLSTM

In the field of information security, steganography and steganalysis play crucial roles. Steganography refers to the technique of hiding secret information within other non-suspicious data, often within images, while steganalysis is the process of detecting and extracting these hidden messages. The ability to accurately identify images that conceal encrypted or hidden data has become increasingly important in today's digital landscape, particularly in areas such as cybersecurity and secure communication. This study focuses on enhancing the detection of such hidden messages by utilizing advanced deep learning techniques, specifically (CNN) and (BiLSTM) networks. This fusion approach involves merging the strengths of both CNNs and BiLSTMs, which leads to better performance compared to using either technique in isolation. The combined model's accuracy is notably higher than that of previous works, as demonstrated in comparisons on trained datasets. To train these models, the study employs Adam’s optimization algorithm, a widely used and efficient method for training deep learning models. To ensure the robustness and reliability of the proposed model, the researchers utilize K-Fold cross-validation with k=2, which allows for better dataset utilization and reduces the risk of overfitting. The proposed techniques were evaluated using three well-known datasets in the steganography and steganalysis community: HILL, BOSSbase 1.01, and providing a solid benchmark for testing the effectiveness of the model. The results indicate that the proposed methodology not only outperforms existing approaches but also provides a highly efficient solution for detecting hidden data in images, with minimal computational overhead.