Convolutional Neural Network Based Image Steganography

Abstract

In cryptography, image steganography is the activity of hiding message into an image by pixel value alteration depending upon the encryption algorithm. However, hiding images in a cover image is a great provocation today. Here Variational Auto-Encoder architecture, a type of artificial neural network is chosen which learns the maximum from the minimum dimension possible. It consists of three parts- encoder, latent space representation(compression), and decoder. It does image concealing by minimizing the loss and removing noise from the image. The secret image is first routinely embedded into the cover image by the encoding network. Then decoding network is utilized to recover the concealed image. Image Steganography using CNN has good compatibility, accuracy, and minimum loss and can be done on dissimilar data types like remote sensing images and aerial images. This model has good training and loss reduction capabilities depending upon its hyperparameters. This model is also used to train other image datasets that are diverse from this ImageNet dataset, such as remote sensing images and aerial images. This model also understands concealing and extraction ensuring secure steganography.