<![CDATA[Steganalysis plays a crucial role in digital forensics by detecting and retrieving hidden information embedded within digital media. Traditional statistical methods, such as Chi-Square and RS-Analysis, are computationally efficient but ineffective against adaptive steganography techniques that minimize detectable distortions. While recent deep learning-based approaches, particularly Convolutional Neural Networks (CNNs), have improved detection accuracy, most rely on single-stream architectures and focus solely on classification, neglecting the recovery of concealed payloads. This study proposes a dual-stage steganalysis framework that integrates a bifurcated CNN for enhanced detection with a genetic algorithm-based extraction pipeline for payload recovery. The bifurcation architecture extends GBRAS-Net by enabling parallel feature learning paths to capture diverse noise patterns, while the extraction module employs chromosome key encoding, Hilbert Curve scrambling, and LZMA compression to reconstruct hidden data. Evaluations on BOSSbase 1.01 and BOWS 2 datasets show that the proposed method achieves an average detection accuracy of 92.53%, outperforming the original GBRAS-Net (89.8%) and other CNN-based models by a statistically significant margin (p < 0.01). Furthermore, the extraction module achieves 100% payload recovery with perfect data integrity verification. The results demonstrate that integrating bifurcated feature learning with robust extraction addresses critical gaps in current steganalysis, offering a practical forensic tool for both detection and reconstruction of hidden information. This approach has significant potential for applications in law enforcement, cybersecurity, and intelligence operations. ]]>
Copyrights © 2026
