Comparative performance assessment of deep learning based image steganography techniques.

Area of Science:

• Computer Science
• Information Security
• Digital Image Processing

Background:

• Data security is a growing concern, with digital images vulnerable to unauthorized access during transmission and storage.
• Traditional image steganography methods face limitations in embedding capacity and image reconstruction quality.
• Deep learning offers potential solutions for enhancing image steganography techniques.

Purpose of the Study:

• To evaluate the performance of V-Net and U-Net++ convolutional neural network (CNN) encoders for image steganography.
• To conduct a comparative analysis of U-Net, V-Net, and U-Net++ architectures in embedding secret images within cover images.
• To design a universal decoder for extracting secret images from steganographic media generated by different architectures.

Main Methods:

• Implementation of V-Net and U-Net++ CNN architectures for image steganography.
• Comparative performance assessment of U-Net, V-Net, and U-Net++ for secret image embedding.
• Development of a standardized decoder for extracting embedded secret images.

Main Results:

• The U-Net architecture exhibited superior performance compared to V-Net and U-Net++.
• U-Net achieved higher embedding capacity for secret images.
• The stego images and reconstructed secret images using U-Net showed better quality.

Conclusions:

• The U-Net architecture is highly effective for image steganography, outperforming V-Net and U-Net++.
• Deep learning-based image steganography, particularly with U-Net, offers a promising solution for secure data transmission.
• The findings suggest U-Net as a preferred choice for achieving high embedding capacity and quality in image steganography.