CNN approach for medical image authentication

Objectives: To propose a Non-blind watermarking based on a Convolutional Neural Network (CNN). Methodology: An iterative learning model is proposed to ensure robustness and imperceptibility of watermarking process. In the first step, Stationary Wavelet Transformation (SWT) and Singular Value Decomposition (SVD) are used for the initial transformation and for embedding. The neural network is used to determine the relationship between host and watermarked image to extract the watermark. Findings: We have implemented our algorithm using Magnetic Resonant Imaging (MRI) and Computerized Tomography (CT), Mammogram and Retinal Images with different attacks and proved to have good robustness with Normalized Correlation coefficient (NC) value of 0.99 and invisibility feature with Peak Signal to Noise Ratio (PSNR) of 43.77 DB. We have compared our method with that of others and it proves to be good in terms of PSNR and NC values. Novelty/Application: This study provides a novel method to train CNN with both watermarked , attacked images and to classify them. Keywords: Medical image authentication; stationary wavelet transform; convolutional neural networks; singular value decomposition