Sphinx

An image-encryption experiment built on chaotic maps, NCA-CML keystreams, and DNA-style transforms.

Overview • Pipeline • Installation • Usage • Testing • Limitations

---

Caution

Sphinx is not a secure cryptography library. It is an educational prototype and code-rehabilitation exercise. For any real confidentiality or integrity requirement, use standard authenticated encryption such as AES-GCM or ChaCha20-Poly1305. This scheme has not been formally analyzed or security-proven.

Overview

Sphinx is a research-inspired image cipher for studying how design choices affect diffusion, damage locality, and failure modes in custom encryption schemes. It uses:

• SHA-256-derived one-time tokens seeded from an image summary vector plus either local randomness (secrets) or an optional quantum source (ANU QRNG API).
• An NCA-CML-inspired keystream generator for deterministic rule selectors, row/column operations, and per-pixel keystreams.
• DNA-style encoding and operations — one of eight encoding rules per channel, followed by DNA addition, subtraction, or XOR.

The current version is a portfolio cleanup of an earlier prototype, focused on reproducibility, cleaner interfaces, and more honest claims about what the code actually does.

Pipeline

flowchart TD
    A[Load RGB image] --> B[Derive 256-bit experiment token\nimage summary + secrets / ANU QRNG]
    B --> C[Split into tiles]
    C --> D[Per-tile token derivation]
    D --> E[NCA-CML rule & keystream generation]
    E --> F[DNA encode RGB channels\none of 8 rules per channel]
    F --> G[Interleave DNA rows\nshuffle columns]
    G --> H[Apply DNA row operations\naddition · subtraction · XOR]
    H --> I[Decode with complementary DNA rules]
    I --> J[Pixel diffusion stage]
    J --> K[Save encrypted tile]
    K --> L{More tiles?}
    L -- yes --> D
    L -- no --> M[Save encrypted image + token]

Loading

Decryption runs the same pipeline in reverse, tile by tile.

Why Tile-Based Processing

The old implementation used a single image-wide permutation. Its failure mode: any localized damage to the ciphertext — a corrupted region, a partial write — was scattered across the entire decrypted image by the inverse permutation.

The current implementation encrypts each tile independently, which means:

• localized ciphertext damage stays localized after decryption,
• damaged regions are easy to visually inspect and isolate,
• the diffusion/locality tradeoff is explicit and configurable via --tile-size.

Smaller tiles contain damage more aggressively. --tile-size 0 runs a single whole-image block, matching the behavior of the old scheme.

Installation

python -m pip install -r requirements.txt

Usage

Encrypt with the default tile size:

python encrypt.py images/m6kuvrsdpy551.png --token-file output/token.json

Encrypt with a smaller tile size for tighter damage containment:

python encrypt.py images/m6kuvrsdpy551.png --tile-size 32 --token-file output/token.json

Encrypt with analysis plots:

python encrypt.py images/m6kuvrsdpy551.png --plot --token-file output/token.json

Decrypt using the saved metadata file:

python decrypt.py encrypted_output/<encrypted-file>.png --token-file output/token.json

Decrypt by passing the token directly:

python decrypt.py encrypted_output/<encrypted-file>.png --token <64-hex-token> --tile-size 64

To use the ANU quantum RNG source, set the SPHINX_ANU_API_KEY or ANU_API_KEY environment variable before running.

Project Structure

sphinx/
├── encrypt.py                        # CLI entrypoint — encryption
├── decrypt.py                        # CLI entrypoint — decryption
└── functions/
    ├── pipeline.py                   # Deterministic tile-local orchestration
    ├── generateSecureKey.py          # Token derivation and analysis helpers
    ├── ncaCml.py                     # NCA-CML parameter updates, rule selection,
    │                                 # permutations, and keystream generation
    └── matrixDNAManipulator.py       # 8 DNA rules + addition, subtraction, XOR, inverses

Testing

python -m unittest discover -s tests

The test suite covers:

• encrypt/decrypt round-trip correctness,
• DNA encode/decode round-trip,
• RGB channel ordering,
• deterministic NCA-CML key-material generation,
• CLI smoke tests,
• localized damage staying within its originating tile.

Limitations

• The algorithm is custom and provides no formal security guarantees. Do not use it to protect sensitive data.
• Tile-local processing improves resilience to partial data loss at the cost of reduced global diffusion relative to a whole-image permutation.
• The implementation is inspired by chaos-based image-encryption literature (see Reference), but is a pragmatic reconstruction, not a claim of exact scientific reproduction.
• The ANU QRNG path is optional and requires an API key.

Reference

Inspired by chaos-based image-encryption literature, including:

Hua, Z. et al. (2018). 2D Sine Logistic modulation map for image encryption. Information Sciences. doi:10.1016/j.ins.2018.04.013