Image Watermarking Based Data Hiding by Discrete Wavelet Transform Quantization Model with Convolutional Generative Adversarial Architectures

Abstract

Traditional watermarking methods can remove a watermark from an image, making it possible to see the copyright information about the image owner or to estimate similarities using techniques such as bit error rate and normalized correlation. Deep learning is another examination field in AI, and is utilized to develop a deep network to extract objective elements and afterwards distinguish the general environment. To assure the robustness and security of computerized image watermarking, we propose a novel algorithm using convolutional generative adversarial neural networks. This research proposed a novel technique in digital watermarking, with data hiding based on segmentation and classification, using deep learning techniques. The used input images are medical images, including Magnetic Resonance Images (MRI) and Computed Tomography (CT) images, which have been processed for noise removal, smoothening and normalization. The processed image has been watermarked using the Singular Value Decomposition-based discrete wavelet transform quantization model, being segmented and classified using convolutional generative adversarial neural networks. The experimental analysis has been carried out in terms of bit error rate, Structural Similarity Index Measure (SSIM), Normalized Cross-Correlation (NCC), training accuracy, and validation accuracy. This achieved an attained bit error rate of 71%, an SSIM of 56%, a Normalized Cross-Correlation of 71%, a training accuracy of 98%, and a validation accuracy of 95%.

Keywords

• digital watermarking
• data hiding
• segmentation
• classification
• deep learning