Encryption Details
SafeLicensing uses a dual-pass chaotic XOR cipher on each detected license plate region.
The logistic map
The logistic map is a classic chaotic dynamical system:
x_{n+1} = r * x_n * (1 - x_n)
Parameters used:
| Parameter | Value | Reason |
|---|---|---|
r | 3.9 | Fully chaotic, non-periodic regime |
x_0 | seed | Initial value in (0.0, 1.0) exclusive |
With r = 3.9, the sequence is extremely sensitive to initial conditions. Changing
seed by 1e-10 produces a completely different key stream after a few hundred iterations.
Encryption pipeline
For each detected plate region (shape (H, W, 3), dtype uint8):
Step 1: XOR with chaotic key
key1 = generate_key(seed, H * W * 3) # chaotic byte sequence
region = region ^ key1.reshape(H, W, 3)
generate_key iterates the logistic map n times, multiplies each x_i by 255,
and casts to uint8.
Step 2: Pixel shuffle
shuffled, indices = shuffle_pixels(region, seed)
All pixels are permuted using a deterministic, seed-derived permutation. The indices are discarded after shuffling; the same seed reconstructs them during decryption.
Step 3: XOR with second chaotic key
second_seed = min(seed * 1.1, 0.9999)
key2 = generate_key(second_seed, H * W * 3)
result = shuffled ^ key2.reshape(H, W, 3)
A perturbed seed (min(seed * 1.1, 0.9999)) generates a second independent key stream.
The cap at 0.9999 prevents the seed from reaching 1.0, which is a fixed point of
the logistic map where all subsequent values collapse to zero.
Determinism
Given the same seed and region dimensions, encryption is fully deterministic:
enc1 = encrypt_image(region, seed=0.42)
enc2 = encrypt_image(region, seed=0.42)
assert np.array_equal(enc1, enc2) # always True
This makes the scheme auditable and reproducible for research.
Seed selection
| Seed | Effect |
|---|---|
| Close to 0 or 1 | Avoid: logistic map behavior degenerates near boundaries |
| 0.1 to 0.9 | Recommended range; all values produce fully chaotic sequences |
| Different per dataset | Change seed between datasets to prevent correlated output |
Security note
This scheme is designed for privacy-preserving anonymisation of license plates in research datasets. It is not a cryptographic cipher and does not provide provable security guarantees against a determined adversary.
For applications requiring strong confidentiality guarantees, use an AES-based approach.
Low-level API
All primitives are directly accessible for research use:
from safelicensing.encryption import logistic_map, generate_key, shuffle_pixels, encrypt_image
# Single logistic map step
x1 = logistic_map(r=3.9, x=0.42)
# Generate a chaotic byte key
key = generate_key(seed=0.42, n=1024)
# Shuffle pixel positions
shuffled, indices = shuffle_pixels(arr, seed=0.42)
# Full two-pass encryption
result = encrypt_image(arr, seed=0.42)
See the Low-level API reference