SVM-SS watermarking model for medical images

Abstract

To ensure the crucial integrity and confidentiality of patients' information, this paper proposes a novel support vector machine (SVM) cum spread spectrum (SS) watermarking model to watermark medical images. In applying watermarking to secure medical images, there are generally three key stages, namely classifying the medical images into Region of Interest (ROI) and Region of Non-Interest (RONI), embedding the patients' information and other relevant information into the image, and lastly extracting that information from the watermark images. The classifying and embedding stages require specific techniques tailored to their different requirements while the third is usually done using symmetric algorithms applied in the embedding stage. Among the soft computing techniques, SVM excels in classification including image classification and has a high potential to be used in the watermarking to improve its performance. However, based on current works reviewed in medical image watermarking, none has applied SVM yet. Similarly, SS is robust to the most common of signal processing and geometric distortions have been successfully applied in image watermarking. Therefore, in our novel model, SVM will be applied in the first stage while SS will be applied in the second and third stages. Significantly, the model aims to secure medical images to resist distortion as well as to avoid medical images quality degradation. The patient confidential data will be embedded into the RONI of their medical images using grayscale JPEG format using the SS symmetric algorithm. The watermark images will be evaluated on robustness and imperceptibility. Experiments will be conducted to measure the similarity ratio (SR) to test the robustness and Peak Signal to Noise Ratio (PSNR) to test the imperceptibility. The results show a high quality robust and imperceptible watermarking has been achieved with SR of more than 0.98 and PSNR of more than 40dB. © 2011 Springer-Verlag.