.png){kind=icon}
.png){kind=icon}
Return
Visually secure image encryption: Exploring deep learning for enhanced robustness and flexibility
.png){kind=icon}
PRE
AI
DOI：10.1016/j.eswa.2024.126027.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
For security reasons, plain images are typically encrypted into noise-like secret images. However, they can easily be detected by potential attackers during transmission or storage, making them vulnerable to interception and corruption. This has led to an increased focus on visually secure image encryption methods. This paper presents a robust and flexible visually secure image encryption method. First, anew compression and reconstruction network based on meta-learning compressed sensing (Meta-learning CS) is devised with the objective of compressing the plain image. Then, the compressed image undergoes encryption through a 2DSLC hyperchaotic map, resulting in the generation of a noise-like secret image. Finally, a visually secure cipher image is obtained using a designed newly embedding and separation network based on traditional deep learning. Experimental results indicate that this method demonstrates superior resilience to advantages in resisting Gaussian noise attack and speckle noise attack relative to other methods. Additionally, the method offers the unique advantage of converting lossless cipher images into JPEG lossy format while maintaining high-quality decrypted images, reducing the risk of drawing attackers' attention and improving storage and transmission efficiency.
Keywords:
Visually secure image encryption
Meta-learning
Compressed sensing
Traditional deep learning
Hyperchaotic map
Journal
IF:
7.5
Papers:
2.9W
Citations:
10.2W