Year: 2015
Type: dissertation
Abstract: Securing medical images is important to protect the privacy of patients and assure data integrity. Consequently, encryption and watermarking were explored to improve the security of medical images. A novel encryption method is introduced to reduce the implementation time and maintain the robustness of algorithms such as AES and 3DES. This method uses selective encryption in an adaptive way. The medical image was divided according to threshold criteria into the region of interest (ROI) and region of background (ROB). If ROB is large, the two-region algorithm is used. The criteria used to determine the threshold values are the K-means or region growing. If ROB is small, multi-region algorithm is used. The multi-region algorithm is a novel adaptive approach based on the variation of information. The threshold values of the regions are obtained using genetic algorithms (GA) tuned with selected parameters. A modified Gold code is designed as an encryption algorithm for background regions. A new algorithm based on the Chinese remainder theorem is modified to encrypt a medium information region. The high information region is encrypted using the AES. The results showed that implementation time is reduced by an average of 20% and the robustness is maintained in most cases. In some cases, the correlation coefficient is high; therefore, an adaptive stopping-criterion permutation algorithm is designed. Non-reversible watermarking is utilized to achieve ownership verification as well as integrity of medical images. To avoid an incorrect diagnosis, the watermark should not modify the informative region. Image segmentation is applied to identify the lowest information region. Embedding maps based on location and entropy are built to reduce cropping attack effect. The DCT domain is used to achieve better invisibility and robustness. GA-based adaptive embedding algorithm is designed to select the frequency coefficients among three DCT models. The GA fitness function is designed using the variance and the SSIM. Invisibility improved to obtain better security for various modalities.
Author: Ahmed Mahmood
Institution: University of Guelph
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Keywords: medical image security, selective encryption, adaptive encryption, region of interest segmentation, K-means clustering, region growing, variation of information, Gold codes, Chinese remainder theorem, AES, image integrity, ownership verification, cropping attacks, DCT domain watermarking, medical image privacy
Open Access status: green
OpenAlex ID: W2547534902
