Contribution of neural networks in image steganography, watermarking and encryption

Abstract

In front of information fishing, it is necessary to protect the information to be circulated over public channel. The goal of this paper is to propose an approach which inserts a logo inside of cover image and having three different possible results according to user's needs. The algorithm can perform steganography, watermarking or host encryption. This technique is based on singular value decomposition (SVD) and discrete wavelet transform (DWT) composed by three steps: pre‐processing, embedding and extraction/decryption. Reference matrix and reference parameter are the two factors deciding the operation and the intervention of neural networks makes the choice of their parameters easier according to the desired result. For measuring the steganography performance, the authors adopt the structural similarity index measure (SSIM) which calculates the similarity between original and watermarked image and peak signal‐to‐noise ratio. For watermarking, the normalized correlation (NC) coefficient investigates the correlation between the original watermark and the extracted watermark. Attacking the watermarked image with common attacks used by previous publication searchers, the value of SSIM and NC coefficient are closed to one. Regarding peak signal‐to‐noise ratio, the overall score is around 61.73 dB. The performance score is not negligible also for the encryption method.

Super-resolution deep neural network (SRDNN) based multi-image steganography for highly secured lossless image transmission

“…Authors Shumeet Baluja et al in Ref. 9 have proposed a model for image steganography. With little distortion to cover images, a N x N x RGB pixel secret image is concealed inside another N x N x RGB pixel cover image.…”