Ali Ghaffari nejad

  abstract

  Mutual information (MI) criterion as a similarity measure is a concept assessing the likeness of two images or probability distribution. Increasing the similarity between to images, by any means, would help us to estimate the total structure and behavior of an image, in case we have the other image in hand. Registration of medical images is one of the common issues in current medical practices, which is developing very rapidly in both new findings and new areas of research. Alignment of functional images, which mostly show soft organs, with structural images, that basically depicts hard parts of the body, is the main objective of the medical image registration. Despite the intrinsic difference between functional and structural images of the same scene, there are still some similarities in image intensity values of the same organ in both images. Similarity between corresponding tissues in two images can be increased by applying a transform to one image. Measuring the amount of similarity among two images is upon to MI. Therefore, we expect that the best alignment between images occur when the likeness of images is the maximum amount according to MI criterion. In this thesis we present an improved algorithm for medical image registration based on MI criterion, along with a new configuration for normalized form of MI that has smoother variations which makes it possible for heuristic algorithms to play a role in registration algorithm.

  Keywords : Image registration, Mutual information, Entropy, Similarity measure

  Date: 22/4/89 Time: 16

  Place: School of Electrical Eng. Seminar Room