Enhanced Steganography Technique via Visual Cryptography and Deep Learning

Abstract

AbstractSteganography is one kind of information hiding technique where a file is hidden within a transferable medium, such as an image, video, or file. Many steganography methods have been proposed and implemented over the decades among which image-steganography is very popular. In image steganography, one of the most popular techniques is the Least Significant Bit (LSB) technique. However, there are certain security drawbacks to this method, such as the fact that anyone who knows where the information is concealed may simply recover it. In this paper, a new approach is proposed by integrating steganographic technique with deep learning and visual cryptography to solve the problem where a secret image can be hidden in a cover picture using steganography but the content of the secret image is embedded by both deep learning and visual cryptography first. The secret picture is initially sent into the autoencoder, which consists of an encoder and a decoder. It compresses the image and renders it unrecognizable. The picture is then subjected to visual cryptography by conducting an exclusive OR (XOR) operation on it with a randomly generated image named mask1. Using the LSB technique, the encrypted secret picture is then concealed within the carrier (cover) image. All of the encoding stages are reversed for the decryption of the secret picture. The consistency of the stego picture was assessed using image quality matrices, putting the experimental findings to test. The values of the image quality metrics indicate the enhancement of security. A comparison study was also conducted with various current tools, and our technique was shown to be superior to the majority of them.KeywordsVisual cryptographyLSBSteganographyAutoencoderDeep learning