README.md

metadata

license: mit
task_categories:
  - image-classification
tags:
  - steganography
  - LSB
pretty_name: Multilingual LSB Stego Dataset
size_categories:
  - 100K<n<1M

Multilingual LSB Stego Dataset

Dataset Summary

This dataset contains 432,000 RGB images (~345 GB) generated to study the detectability of Least Significant Bit (LSB) steganography under controlled, reproducible settings. It enables systematic analysis across multiple resolutions, bit-depths, payload ratios, embedding policies, and languages, and includes per-file metadata for deterministic message recovery.

• Total size: ~345 GB
• Images: 432,000
• Languages (payload): EN, DE, ES, FR, IT, TR
• Resolutions: 128×128, 256×256, 512×512, 1024×1024
• Bits per byte (LSB depth): 1, 2, 4
• Payload ratios (of capacity): 0.25, 0.50, ~1.00
• Embedding policies: sequential, random (seeded)
• Metadata: JSON for each stego image (original filename, message length, bpp, policy, seed, status)
• Deterministic header: flag | msg_len(32) | num_bits_code(2) | [seed(32)] for exact extraction

Abstract

TBA

Folder Structure

LSBStegoDataset/
 ├── 128x128_en/      256x256_en/      512x512_en/      1024x1024_en/
 ├── 128x128_de/      256x256_de/      512x512_de/      1024x1024_de/
 ├── 128x128_es/      256x256_es/      512x512_es/      1024x1024_es/
 ├── 128x128_fr/      256x256_fr/      512x512_fr/      1024x1024_fr/
 ├── 128x128_it/      256x256_it/      512x512_it/      1024x1024_it/
 └── 128x128_tr/      256x256_tr/      512x512_tr/      1024x1024_tr/

Each leaf folder contains:

• Stego images (as PNG)
• JSON metadata for each stego image (embedding parameters & verification)
• (Optionally) histogram PNG/JSON if generated

Usage

from datasets import load_dataset

# Load entire repository
ds = load_dataset("emirozturk/LSBStegoDataset")

# Load a specific subset/folder (example: 256x256 Turkish)
ds_tr256 = load_dataset(
    "emirozturk/LSBStegoDataset",
    data_files="256x256_tr/*"
)

Typical Research Tasks

• Detectability vs. payload × bpp × policy × language × resolution
• Cross-lingual generalization of steganalysis models
• Feature-level analyses of language-induced bit statistics

Citation

If you use this dataset, please cite:

TBA

License

See the repository license for terms of use.