Skip to content

deepbrainscu/Image_Steganalysis

Repository files navigation

Image Steganalysis

GA-STEGNET: Görüntü Steganalizinde Genetik Algoritma Tabanlı Kolektif Derin Öğrenme Modeli
GA-STEGNET: A Genetic Algorithm-Based Ensemble Deep Learning Model for Image Steganalysis


TR – Özet

Görüntü steganalizi, sayısal görüntüler içerisine gömülen gizli mesajları tespit etmeyi hedefleyen ve siber güvenlik açısından kritik öneme sahip bir tekniktir. Geleneksel yöntemlerin sınırlı genelleme kabiliyetleri, derin öğrenme tabanlı yaklaşımları bu alanda daha etkin bir çözüm haline getirmiştir. Bu çalışmada, literatürde kabul görmüş derin öğrenme modelleri mimari modifikasyonlarla optimize edilmiş; bu modellerin softmax çıktıları, Genetik Algoritma (GA) tabanlı adaptif ağırlık katsayılarıyla birleştirilerek GA-STEGNET adlı kolektif bir model geliştirilmiştir. Modelin performansı, farklı uzamsal steganografi yöntemleri altında 0,2 ve 0,4 bpp veri gizleme oranları kullanılarak değerlendirilmiştir. Deneysel sonuçlar, GA-STEGNET'in 0,4 bpp senaryosunda %78,1–%87,6 ve 0,2 bpp senaryosunda %65,9–%78,5 doğruluk aralığına ulaştığını göstermiştir. Önerilen yaklaşım, az sayıda model kombinasyonuyla dahi hem bireysel modellerden hem de standart topluluk yöntemlerinden daha yüksek ve daha kararlı bir steganaliz başarısı sağlamıştır.

EN – Abstract

Image steganalysis aims to detect hidden messages embedded in digital images and is critical for cybersecurity. The limited generalization of traditional methods has made deep learning-based approaches more effective in this domain. In this work, established deep learning models are optimized via architectural modifications; their softmax outputs are combined with Genetic Algorithm (GA)-based adaptive weights to form GA-STEGNET, a collective ensemble model. Performance is evaluated under various spatial steganography methods at 0.2 and 0.4 bpp (bits per pixel). Results show GA-STEGNET achieves 78.1–87.6% accuracy at 0.4 bpp and 65.9–78.5% at 0.2 bpp. The approach yields higher and more stable steganalysis performance than both individual models and standard ensemble methods, even with few model combinations.


Proje Yapısı / Project Structure

Image_Steganalysis/
├── utils/                     # Network, Data_Generator, data_handling, metric_utils
├── Ensemble_Models/           # Trained model weights (.h5, .hdf5)
├── Dataset_samples/           # Test samples
├── predict_steganalysis.py    # Single image prediction (Cover/Stego)
├── image_stegnalysis_train.py
├── image_stegnalysis_ensemble.py
└── image_stegnalysis_train_genetic_ensemble.py

Kurulum / Installation

pip install tensorflow numpy pandas opencv-python scikit-learn scipy openpyxl deap tensorflow-addons

Kullanım / Usage

Tahmin / Prediction – Set IMAGE_PATH, STEG_METHOD, BPP_RATE in predict_steganalysis.py:

python predict_steganalysis.py

Eğitim / Trainingimage_stegnalysis_train.py or image_stegnalysis_train_genetic_ensemble.py

Modeller / Models

QIANNET, XUNET, YENET, YEDNET, SRMNET, GBRASNET – HUGO, MIPOD, WOW, SUNIWARD at BPP 0.2 and 0.4.

Atıf / Citation

GA-STEGNET: Görüntü Steganalizinde Genetik Algoritma Tabanlı Kolektif Derin Öğrenme Modeli. Gazi Üniversitesi Mühendislik ve Mimarlık Fakültesi Dergisi, 2026.

GA-STEGNET: A Genetic Algorithm-Based Ensemble Deep Learning Model for Image Steganalysis. Journal of Faculty of Engineering and Architecture of Gazi University, 2026.

Referans / Reference

Network architectures adapted from BioAITeam/Steganalysis.

About

No description, website, or topics provided.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors

Languages