IMAGE STEGANOGRAPHY WITH CNN BASED ENCODER-DECODER MODEL-DL

Abstract

Image steganography, the technique of hiding secret information within an image, has gained significant importance because of its applications in secure communication. Traditional steganography methods usually depend on simple pixel transformations, which can be easily detected by modern steganalysis techniques. This project proposes a new approach to image steganography using a Convolutional Neural Network (CNN) based Encoder–Decoder model. The proposed system uses deep learning methods to embed hidden messages into the least significant bits (LSBs) of an image while training an encoder–decoder architecture to recover both the original image and the concealed message. The CNN-based encoder–decoder model is designed to improve the invisibility of the hidden data and increase robustness against common image operations such as compression or resizing. The encoder extracts important features from the input image and embeds the secret message, while the decoder reconstructs the stego-image and retrieves the hidden information. To maintain the quality of the stego-image, a loss function is applied to balance the visual quality of the image and the accuracy of message extraction.