ABSTRACT

This paper investigates the best possible optimisation strategy for determining global extremes of objective functions in the framework of image registration. We register serial optical slices captured by a confocal laser-scanning microscope (CLSM). We experiment with following objective functions: the Sum of Absolute Valued Differences (SAVD), Normalised Correlation Coefficient (NCC) and Mutual Information (MI). Full search of the parametric space for the global extreme is tedious, therefore, we study following optimisation strategies: the genetic algorithm, downhill simplex method, Powell’s method, and simulated annealing, all with the use of multiresolution approach to decrease the computational load.