A Fast Medical Image Registration Based on Steerable Pyramid

In this paper, we present a fast algorithm for medical image registration. Through utilizing steerable filters, we can extract multi-directional feature information of image steadily and then combine it to obtain synthesized feature graph for more accurate registration. To avoid a large amount of computation on globally searching rotation center, we present a method of fixing centroid of feature graph as rotation center for making geometry transform and using steerable pyramid for level by level registration. Furthermore, calculating centroid can also provide initial estimation of translation parameters, which greatly accelerates search time. Simulation results prove that the accuracy of the algorithm is high and the computing time of the algorithm can be remarkably reduced. Applying the algorithm to the multimodal image registration also has a good effect.