CNN – Based GIF Steganography Using Lightweight U-Net

Abstract

This paper introduces a novel approach to image steganography by leveraging Convolutional Neural Networks (CNNs) to embed secret data into `.gif` file formats. Traditional steganographic techniques have largely focused on static image formats such as JPEG and PNG, neglecting the widely used GIF format. GIFs, with their palette-based compression and multi-frame structure, offer unique opportunities and challenges for steganographic embedding. We propose a lightweight U-Net variant based on the U-Net design, which comprises a hiding network to embed secret data and a revealing network to extract it. Our model is evaluated on GIF-based datasets, and performance is assessed using standard metrics such as PSNR, SSIM, and Bit Error Rate (BER). Experimental results demonstrate high fidelity, imperceptibility, and robust message recovery, positioning our method as a promising direction for secure and covert communication in modern multimedia formats.