Optimized Rotation-Invariant Coordinate Convolutional Neural Network based Digital Image Watermarking Technique

Abstract

The rapid growth of multimedia and network technology, accessing digital media has become increased easily. Watermarking techniques are essential for protecting digital images.  Consequently, protecting intellectual property has heightened the need for effective image watermarking. Although various image watermarking approaches have been developed to address this need, they often face challenges related to robustness and transparency. In this manuscript, Optimized Rotation-Invariant Coordinate Convolutional Neural Network Based Digital Image Watermarking Technique (DIW-RICCNN-HBO) is proposed. Initially the images are collected from CIFAR10 and Pascal VOC2012 dataset. Then the collected images are preprocessing by Adaptive Variational Bayesian Filter (AVBF) to remove noise. Then the preprocessed images are embedding the cover image containing the secret data of the embedding network. After that Design an encoder network using Rotation-invariant coordinate convolutional neural network (RICCNN) to extract hidden features via both the cover images and the secret mark images. The Honey Badger Optimization (HBO) method is proposed to optimize the weight parameters of the RICCNN for improved results in digital image watermarking. The proposed DIW-RICCNN-HBO method is implemented on Python. The proposed method attains 35.66%, 32.73%, and 31.43% higher PSNR and 32.77%, and 28.93% higher SSIM comparing with existing techniques like Water chaotic fruit fly optimization-based deep convolutional neural network for image watermarking using wavelet transform (DIW-DCNN), A Robust Document Image Watermarking Scheme using Deep Neural Network (DIW-DNN) and ReDMark: Framework for residual diffusion watermarking based on deep networks (RDW-ReDMark) methods respectively.