A Robust and Secure High-Capacity Coverless Information Hiding Scheme Using Deep Learning

Abstract

Coverless image steganography hides secret information without directly changing the cover image. Because the original image is not modified, it is more difficult for image steganalysis tools to detect the hidden information. In this research, Wasserstein Generative Adversarial Networks (WGAN) and Local Binary Pattern (LBP) features are used to securely hide and transmit sensitive information using a single cover image. The proposed method provides higher hiding capacity and better security than traditional methods that require multiple cover images.First, the cover image is divided into overlapping blocks. The LBP feature is calculated for each block and converted into a hash code. The secret message is then matched with image blocks having the same hash codes. A lookup table is used to make the embedding process faster.The generated stego information is given to a GAN model to create a meaningful but unrelated image. This image can be sent to the receiver instead of directly sending the stego image. At the receiver side, the WGAN model is used to reconstruct the required stego information. The use of overlapping blocks helps generate many unique hash codes. Without overlapping blocks, a single image may not provide enough hash codes to hide the complete secret message.