Color image copyright protection based on watermarking using machine learning in the wavelet-packets domain

Abstract

Abstract Image watermarking technologies have been developed to meet the concerns in multimedia copyright protection caused by the tremendous demand on multimedia sharing. In this work, a novel color image watermarking scheme is proposed for copyright protection applications. The developed scheme employs the wavelet packets decomposition (WPD) along with utilizing the back-propagation feed-forward neural networks (BPFNN). Here, a BPFNN-based color image watermarking using WPD decomposition approach is proposed and is denoted as FNN-WPD. The produced WPD-packets, from each of the RGB color channels, are reordered into chains having specific spatial- and frequency-based features. These chains are used for mark embedding and extraction using the BPFNN construction. Simulations have shown superior imperceptibility (PSNR ≥ 60 dB) and improved high robustness against incident and malicious attacks, while exhibiting low computational cost, especially at the extraction stage. Moreover, the scheme has been extended to be used for QR-code embedding with very promising robustness performance.