Research Of Algorithm Of Medical Image Segmentation And Registration

Image segmentation and registration have been two important tasks in image processing. In recent decades, although researchers proposed many new methods for segmentation and registration, none of these methods have a wide range of versatility and there are many problems in practical applications. The segmentation and registration, therefore, will remain the focus of research in medical image processing. In this thesis, the main research primarily focuses on these two image processing technology. The main works are as follows:1. A novel algorithm formulated by add an energy term based on Fisher criterion to the objective function of LBF model. We can avoid the level set curve fall into local minimum during evolution by introduce global statistics of pixels on both sides of the curve. Such improvement reduces the dependence of segmentation results to the initial position of contour and enhances robustness and practicality of the algorithm.2. The traditional mutual information registration method has the disadvantages of slow speed of convergence and large amount of calculation. A coarse-to-fine mutual information registration method that adopts improved Powell algorithm could improve these disadvantages to a great extent. The improved Powell algorithm can avoid converge to local optimal solution in the process of search for the transformation parameters. The registration results of rough registration to small resolution image can be used as the initial search point of original image which can speed up the search speed and reduce the time of registration.3. In order to deal with the registration of lager deformation or displacement, this thesis studies a optical flow registration method which based on Gaussian pyramid and Lucas-Kanade algorithm. This method makes difference between two images becomes sufficiently small by pyramiding the original images. The effectiveness of this registration algorithm be proved the through experiments.