Combining quantum-inspired keying, AI-driven smart compression, and noise-adaptive embedding for robust, low-detectability covert channels

A system to hide digital messages across media carriers (images/files) that increases capacity, security, and robustness by blending cryptographic primitives, machine learning compression, and adaptive embedding. Core components:

Quantum-inspired key generator

High-entropy pseudo-random key material from quantum-inspired state sampling; keys control fragmentation, permutation, and per-carrier encryption.

Multi-carrier splitting

Message segmented and distributed across multiple carriers to reduce exposure and enable partial recovery if a carrier is lost.

AI-based smart compression

Trained autoencoder to produce compact, recoverable latents — increases effective payload capacity while preserving retrievability.

Noise-adaptive embedding

Local noise/SNR estimator tunes embedding strength (α) per-region for better imperceptibility and retrieval under distortion.

Flexible embedding engine

Supports spatial methods (enhanced LSB) and frequency-domain embedding (DCT/FFT); selection driven by robustness/visibility tradeoffs.

Error correction & reconstruction

ECC (e.g., Reed-Solomon or LDPC) applied to compressed payload; reconstruction pipeline reassembles fragments using key-driven permutations.

Losses & training

Composite loss: perceptual/image loss (MSE/Perceptual) + message loss (BCE/BER-aware). Train on datasets augmented with JPEG, noise, rescaling, etc.

Evaluation metrics

PSNR / SSIM for carrier quality; BER and F1 for payload recovery; adversarial tests: JPEG compression, Gaussian noise, rescaling, and steganalysis tools.

Implementation stack

PyTorch or TensorFlow for ML components; optional quantum simulators (Qiskit) for keygen; GPU acceleration recommended.

Operational & security notes

Tunable knobs: compression ratio, α, number of carriers. Trade-off between capacity, detectability, and compute cost. Distribution + encryption minimize full-message leakage if a subset of carriers is compromised