Optimizing Image Steganography: A Comparative Study of Pre-Processing Techniques and the Hybrid U-Encoder Network

Image steganography requires balancing high fidelity, security, and imperceptibility when embedding secret information into cover images. Effective pre-processing and advanced deep learning models are essential for achieving these goals. In this study, we propose an enhanced image steganography framework that combines optimal pre-processing techniques with advanced deep learning-based models to achieve secure, high-quality image embedding. We first evaluate five distinct pre-processing methods-CLAHE, Bilateral Filtering, Gaussian Filtering, Median Filtering, and Laplacian Filtering-on cover images to determine their impact on performance metrics, including Mean Squared Error (MSE), Root Mean Squared Error (RMSE), Peak Signal-to-Noise Ratio (PSNR), Structural Similarity Index (SSIM), and Normalized Cross-Correlation (NCC). Among these techniques, Bilateral Filtering demonstrated superior results with the lowest MSE (58.78), highest PSNR (30.44 dB), and near-perfect SSIM (0.948) and NCC (0.988), making it the optimal pre-processing method for steganography. Following pre-processing, two steganography models were implemented and compared: a simple concatenation-based model and a novel Hybrid U-Encoder Network inspired by U-Net architecture. The Hybrid U-Encoder Network was designed to enhance the embedding and extraction process of secret images. Comparative analysis using MSE, PSNR, and SSIM showed that the Hybrid U-Encoder Network outperformed the concatenation-based model, achieving a lower MSE of 71.55, PSNR of 58.62 dB, and SSIM of 0.982. This research highlights the importance of effective pre-processing in improving the steganography process and demonstrates the advantage of using a Hybrid U-Encoder Network for embedding secret information. By combining Bilateral Filtering with the Hybrid U-Encoder Network, the proposed approach provides a robust solution for secure and imperceptible image steganography, significantly improving the trade-off between embedding capacity and image quality.