Robust watermarking for magnetic resonance images with automatic region of interest detection

Abstract

Medical image watermarking requires special considerations compared to ordinary watermarking methods. The first issue is the detection of an important area of the image called the Region of Interest (ROI) prior to starting the watermarking process. Most existing ROI detection procedures use manual-based methods, while in automated methods the robustness against intentional or unintentional attacks has not been considered extensively. The second issue is the robustness of the embedded watermark against different attacks. A common drawback of existing watermarking methods is their weakness against salt and pepper noise. The research carried out in this thesis addresses these issues of having automatic ROI detection for magnetic resonance images that are robust against attacks particularly the salt and pepper noise and designing a new watermarking method that can withstand high density salt and pepper noise. In the ROI detection part, combinations of several algorithms such as morphological reconstruction, adaptive thresholding and labelling are utilized. The noise-filtering algorithm and window size correction block are then introduced for further enhancement. The performance of the proposed ROI detection is evaluated by computing the Comparative Accuracy (CA). In the watermarking part, a combination of spatial method, channel coding and noise filtering schemes are used to increase the robustness against salt and pepper noise. The quality of watermarked image is evaluated using Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index (SSIM), and the accuracy of the extracted watermark is assessed in terms of Bit Error Rate (BER). Based on experiments, the CA under eight different attacks (speckle noise, average filter, median filter, Wiener filter, Gaussian filter, sharpening filter, motion, and salt and pepper noise) is between 97.8% and 100%. The CA under different densities of salt and pepper noise (10%-90%) is in the range of 75.13% to 98.99%. In the watermarking part, the performance of the proposed method under different densities of salt and pepper noise measured by total PSNR, ROI PSNR, total SSIM and ROI SSIM has improved in the ranges of 3.48-23.03 (dB), 3.5-23.05 (dB), 0-0.4620 and 0-0.5335 to 21.75-42.08 (dB), 20.55-40.83 (dB), 0.5775-0.8874 and 0.4104-0.9742 respectively. In addition, the BER is reduced to the range of 0.02% to 41.7%. To conclude, the proposed method has managed to significantly improve the performance of existing medical image watermarking methods.