Abstract
In the research of life science, a fruit fly, Drosophila melanogaster, with the abilities of learning and memory is chosen for research to facilitate the understanding of structures and functions of the brain neural network. In fruit flies’ brains, the mushroom bodies are important neuropils, known to be involved in learning and memory. Due to the importance of mushroom bodies and its characteristic shape, registration for the mushroom bodies is an interesting and useful research topic. The system we developed in this thesis provides an automatic registration applied to mushroom bodies. We register the standard model of mushroom bodies to individual data set of confocal images with noises.A mainly two-stage surface-based registration algorithm has been proposed in this study. This algorithm is composed of global and local registration. Affine registration has been chosen as the global registration to eliminate the large, global differences such as scale and orientation between the two surface models. And a two-level local registration has been proposed to eliminate the remaining differences between the surfaces after the global registration. The two-level local registration initially first eliminates the angular differences and the length differences of the six axes between the target model and the standard model. Then, performing the point-based warping eliminates finer variations between the two models. The experimental results show that performing the registration algorithm proposed here makes the volume data registered satisfactorily roughly well.