High-Capacity Image Steganography Based on Improved Xception

Area of Science:

• Computer Science
• Information Security
• Artificial Intelligence

Background:

• Traditional cover modification steganography methods suffer from limited data hiding capacity.
• Deep learning approaches offer potential for enhanced steganography capabilities.

Purpose of the Study:

• To develop a high-capacity image steganography method using the Xception convolutional neural network architecture.
• To improve steganography speed and embedding efficiency compared to existing methods.

Main Methods:

• Utilized the Xception architecture, incorporating deep separable convolutional layers and jump connections for improved gradient flow and multi-scale information processing.
• Developed a novel steganography process involving cascading secret and cover images, embedding data via a hidden network, and reconstructing the secret image through an extraction network.

Main Results:

• The proposed method achieves a high average embedding capacity of 23.96 bits per pixel (bpp).
• Reconstructed images exhibit high peak signal-to-noise ratio (PSNR) and structural similarity (SSIM), indicating excellent visual quality.
• Demonstrated significantly improved steganography speed.

Conclusions:

• The Xception-based steganography method effectively addresses the limitations of traditional techniques, enabling large-capacity, high-quality, and efficient data hiding.
• This novel approach represents a significant advancement in the field of image steganography.