Medical Image Registration can be seen as an optimization problem, maximizing or minimizing a cost function by optimization techniques. In this paper, we evaluate the performances of different kinds of optimization methods, such as Powell's method, downhill simplex, gradient descent, regular step gradient descent, conjugate gradient descent and simulated annealing, for rigid medical image registration. Cost function is based on mutual information and all of the optimization methods are cooperated with a multiresolution strategy. Experiments show that gradient-based methods are better than non-gradient based methods and they are superior when the cost function is biased by some noise.
展开▼
