PARALLELIZATION FOR A NON-RIGID REGISTRATION USING ELASTIC MODEL, FINITE ELEMENT METHOD AND MUTUAL INFORMATION

摘要

The aim is to parallelize an approach for non-rigid registration. We have led a survey on a new approach in non rigid registration using elastic model, finite element method and mutual information and we have had good results. Using a deformation algorithm, based on the isotropic linear elastic material model, one can get a good non-rigid registration result. This model allows to get image regularization characteristics. However, the Mattes et al. mutual information [1] is reputed to do a multimodal registration. Besides, one can use it for a registration with images from the same modality but in low resolutions, such as the Positron Emission Tomography (PET). This approach concerns a process with a fully automatic registration. However one can note that registration requires a too much important calculation time and high occupancy spaces especially with 3D images. Thus, the parallelization is required. So, this paper deals the transformation of the registration algorithm in a parallel environment. We have worked with Single Program Multiple Data model on Distributed Memory (SPMD-DM) architecture.