Image Steganography Using U-Net Architecture for Seamless Payload Embedding and Extraction

Image steganography embeds hidden messages within digital images so they remain imperceptible through a process known as hiding. Research investigates how the U-Net architecture implements seamless operations for data-embedded payload images while preserving data security and visual integrity. Current image steganography techniques that utilize Least Significant Bit and transform domain approaches have limited capacity for embedding data, while causing visual artifacts and struggling against noise and compression artifacts. We overcome the current challenges by implementing a Deep Convolutional Encoder-Decoder Network modelled after the U-Net architecture that uses skip connections to preserve spatial characteristics while learning efficient payload retrieval. Through its embedding approach, the proposed method takes secret images that become hidden within cover images to produce stego images that maintain the original visual appearance until the subsequent extraction process recovers the secret data with high accuracy. End-to-end training of this model utilises a multi-factor loss function to generate stego images with minimum distortion, coupled with precise payload extraction performance. Results from experimental trials demonstrate that the U-Net steganography system achieves better PSNR, along with higher SSIM values and extraction precision, compared to traditional steganography approaches.

Ҳали таржима қилинмаган