Inspiration

In a world of digital communication, information can be hidden anywhere - even in plain sight. Steganography, the art of concealing messages within digital images, poses significant challenges for cybersecurity and digital forensics. Our project was inspired by the growing need to detect these hidden communications, protecting sensitive information and understanding the sophisticated methods of digital concealment.

What it does

Our steganalysis model is a sophisticated image analysis tool that: Detects hidden messages within digital images Analyzes pixel-level variations Provides a probability score of steganographic content Can process single images or entire directories Identifies images with potential hidden information with 95% accuracy

How we built it

Used the Alaska2 Image Steganalysis dataset Collected images with three steganography methods (JMiPOD, JUNIWARD, UERD)

Feature Engineering

Developed advanced feature extraction techniques Created a 150-dimensional feature vector Analyzed statistical, gradient, and least significant bit variations

Model Development

Explored multiple machine learning and computer vision algorithms Implemented Random Forest Classifier Optimized feature selection and model parameters

Model Validation

Used stratified cross-validation Implemented rigorous performance metrics Fine-tuned detection thresholds

Challenges we ran into

High Dimensional Feature Space

Managing complex feature extraction Reducing computational complexity longer training hours and high memory usage while exploring pre-trained computer vision models

Accomplishments that we're proud of

Achieved 95% accuracy in detecting hidden messages Developed a flexible, scalable steganalysis model Created a comprehensive feature extraction pipeline Implemented both single and batch image analysis Demonstrated practical applications in cybersecurity

What we learned

Technical Insights Advanced image processing techniques Machine learning model optimization Feature engineering strategies

Built With

• deep-learning
• flask
• machine-learning
• python