Optimization-driven steganographic system based on fused maps and blowfish encryption.

Area of Science:

• Digital image processing
• Information security
• Computer vision

Background:

• Steganography conceals data in images, but balancing imperceptibility, capacity, and robustness against advanced steganalysis is challenging.
• Deep Learning (DL) methods pose a significant threat to traditional steganographic techniques.

Purpose of the Study:

• To propose a novel adaptive image steganography framework.
• To enhance security and visual quality while maintaining high data embedding capacity.
• To ensure resistance against statistical and DL-based steganalysis.

Main Methods:

• Utilized BOSSbase and USC-SIPI datasets for cover and secret images.
• Implemented Blowfish cipher encryption for secret data confidentiality.
• Generated a fused map using entropy and Laplacian noise maps for optimal embedding areas.
• Employed Particle Swarm Optimization (PSO) for embedding location selection and ordering.
• Applied priority-guided Least Significant Bits (LSB) substitution for data embedding.

Main Results:

• Achieved high Peak Signal-to-Noise Ratio (PSNR) of 51-61 dB and Structural Similarity Index Measure (SSIM) of 0.9972-1.0.
• Demonstrated perfect secret reconstruction (SSIM = 1.000) with embedding capacities up to 0.1-1.0 Bits Per Pixel (BPP).
• Security tests showed undetectable steganography with Area Under Curve (AUC) of 0.49-0.57 and Regular Singular (RS) statistics of 0.59-0.66.
• Embedding and extraction processing times were under 0.25 seconds.

Conclusions:

• The proposed adaptive steganography framework effectively balances capacity, visual quality, and security.
• The system demonstrates robustness against sophisticated DL-based steganalysis, including Convolutional Neural Networks (CNN).
• The method offers a practical solution for secure data hiding in digital images.