Hugging Face's logo

Datasets:
Rinovative
/
pd12m_dct_based_synthetic_stegano

Modalities:
Image
Formats:
imagefolder
Size:
1K - 10K
Tags:
pd12m
Synthetic
steganography
dct
computer-vision
image-perturbation
Libraries:
Datasets
License:
Dataset card Data Studio Files
xet
pd12m_dct_based_synthetic_stegano / README.md
Rino Albertin
feat(data): use conseal for dataset generation instead of manual perturbation
69642b2
preview code
|
raw
history blame
3.36 kB
metadata
license: cc0-1.0
tags:
  - pd12m
  - synthetic
  - steganography
  - dct
  - computer-vision
  - image-perturbation
  - public-domain
pretty_name: PD12M Synthetic Steganography (DCT-based)
size_categories:
  - 1K<n<10K
dataset_info:
  source: https://source.plus/pd12m?size=n_100_n
  generated_by: Rinovative (2025)
  description: >
    A synthetically generated steganography dataset based on Public Domain 12M
    (PD12M), simulating three DCT-domain perturbation patterns inspired by
    common steganographic algorithms (JMiPOD, JUNIWARD, UERD). All variants are
    generated from CC0-licensed images using the conseal simulation library.

PD12M DCT-Based Synthetic Steganography Dataset

This dataset is a synthetically generated steganographic image dataset based on the PD12M (Public Domain 12M) image collection.
It simulates detectable modifications produced by real-world JPEG steganography algorithms, using only public domain data.

Each original image is duplicated into three synthetic stego variants, inspired by real-world JPEG steganographic algorithms:

  • synthetic_JMiPOD: simulated using conseal.nsF5, as JMiPOD is not available in conseal
  • synthetic_JUNIWARD: generated via conseal.juniward
  • synthetic_UERD: generated via conseal.uerd

All variants were created using the official embedding simulators of the conseal library.
The embedding strength was set to alpha = 0.4.

Each Stego image contains a simulated payload embedded using conseal.
The payload is not retained or accessible and serves only to produce realistic DCT modifications.

📂 Dataset Structure 

/Cover/                 → Original PD12M images (resized + JPEG recompressed)
/synthetic_JMiPOD/      → Cost-based embedding (via nsF5)
/synthetic_JUNIWARD/    → Texture-aware HF embedding
/synthetic_UERD/        → Uniform random embedding

All images are JPEG-compressed and named using a uniform index (e.g., 00001.jpg).

📜 License and Sources

📚 Citation

If you use this dataset, please reference:

  • The original PD12M image dataset (Public Domain)
  • This synthetic modification (Rinovative, 2025)
  • The conseal simulation library

    Lorch, B., & Benes, M. (2024). conseal: Simulation framework for JPEG steganography. University of Innsbruck.
    GitHub repository: uibk-uncover/conseal

📄 Project Reference

Note:
This synthetic dataset was created as a publicly available demonstration subset for the project:
"Deep Learning for Steganalysis – ALASKA2 Dataset"
(GitHub Repository – Rinovative/alaska2-steganalysis)
(BSc Systems Engineering, Computational Engineering Major, OST – Spring 2025, Rino Albertin).

The project primarily uses the ALASKA2 dataset.
In cases where ALASKA2 is not available, this synthetic dataset provides a compatible fallback for running the full pipeline.

Important:
This demo subset is fixed and remains consistent with the final project version.