LVGG-IE: A Novel Lightweight VGG-Based Image Encryption Scheme.

Area of Science:

• Computer Science
• Artificial Intelligence
• Cryptography

Background:

• Image security is increasingly challenged by AI and computer science advancements.
• Existing chaotic and DNA-based image encryption methods face complexity and degradation issues.
• Deep learning integration in image encryption is nascent with several limitations.

Purpose of the Study:

• To propose a novel, secure, and efficient image encryption scheme.
• To address the security degradation in chaotic systems and complexity in existing methods.
• To leverage deep learning for improved image protection.

Main Methods:

• Developed a lightweight VGG (LVGG) network for key seed generation.
• Integrated the key seed with a chaotic system for plaintext-dependent key generation.
• Employed a dynamic substitution box (S-box) and a single-connected (SC) layer with VGG convolution for image scrambling and encryption.

Main Results:

• Achieved a correlation coefficient between adjacent pixels of 10-4.
• Demonstrated high Number of Pixel Change Rate (NPCR) exceeding 0.9958.
• Confirmed that the Uniformity of Pixel Apperture Change (UACI) falls within theoretical values.

Conclusions:

• The proposed LVGG-IE scheme offers high security, efficiency, and robustness for image encryption.
• The method effectively overcomes limitations of existing chaotic and deep learning-based approaches.
• The scheme is suitable for various applications requiring secure image protection.