Sterilization of image steganography using self-supervised convolutional neural network.

Area of Science:

• Computer Science
• Cybersecurity
• Artificial Intelligence

Background:

• Steganography enables covert communication, increasing network security risks.
• Current steganography sterilization methods require known cover-stego image pairs, limiting effectiveness on unknown images.
• Existing techniques struggle to block stego images effectively in social networks.

Purpose of the Study:

• To propose an effective image steganography sterilization method.
• To address the limitations of existing methods in sanitizing unknown images.
• To enhance the stego image blocking rate in social networks.

Main Methods:

• A self-supervised convolutional neural network (SS-Net) is proposed, eliminating the need for prior steganography knowledge.
• SS-Net utilizes a purification module with pixel-shuffle down-sampling for self-supervised learning.
• A refinement module, incorporating centrally masked convolutions and dilated convolution residual blocks, eliminates secret messages and preserves image quality.

Main Results:

• SS-Net achieves a 100% blocking rate for stego images on BOSSbase test sets.
• The method effectively balances secret message destruction with image quality preservation.
• SS-Net outperforms state-of-the-art methods in message elimination and image quality retention.

Conclusions:

• The proposed SS-Net method provides an effective solution for sterilizing unknown stego images.
• This approach significantly enhances network security by blocking covert communications.
• SS-Net offers superior performance in both message removal and image quality preservation compared to existing methods.