Related Experiment Video
Updated: Apr 4, 2026
03:31
Author Spotlight: Enhancement of Salient Object Detection for Smart Grid Applications
Published on: December 15, 2023
1.2K
Enhanced image encryption with deep generative models using a self-attention mechanism
Ilham Karmouni1, Nawal El Ghouate1, Mohamed Amine Tahiri2
1Engineering, Systems and Applications Laboratory, National School of Applied Sciences, Sidi Mohamed Ben Abdellah University, Fez, Morocco.
Scientific Reports
|April 2, 2026
Summary
This study introduces a novel deep learning system for secure image encryption. The model enhances security and accuracy for medical images, improving visual data protection.
Area of Science:
- Computer Science
- Artificial Intelligence
- Cryptography
Background:
- Digital image security is crucial across various applications.
- Existing encryption methods face challenges in speed and security.
- Deep generative models offer potential for advanced image encryption.
Purpose of the Study:
- To develop a novel image encryption system using deep generative models and self-attention.
- To enhance both encryption and decryption speeds.
- To improve the security and reconstruction accuracy of encrypted images.
Main Methods:
- Utilized CycleGAN-based models for image encryption and decryption.
- Incorporated a self-attention module to capture global image dependencies.
- Evaluated performance on color image datasets, brain MRI, and skin cancer images.
Main Results:
- Achieved high security with Entropy ≈ 7.9996 and NPCR ≈ 99.99%.
- Demonstrated high reconstruction accuracy (SSIM ≈ 0.99, PSNR > 40 dB).
- Showcased strong resistance to differential (UACI ≈ 33.46) and occlusion attacks.
Conclusions:
- The proposed system offers a secure and efficient solution for image encryption.
- Deep generative models with self-attention present promising avenues for visual cryptography.
- The system's effectiveness is validated on critical medical imaging datasets.