Non-rigid motion estimation based on fuzzy models

Image matching of deformable structures has captured great attention in image processing, and specially in the medical field. This paper proposes a method that faces the ill-posed nature of this problem, by using a cluster-sized similarity cost function, the ambiguity in each similarity map is described by a fuzzy parametric model, and, finally, a spatially non-uniform fuzzy interpolation is used to translate the parametric info into a set of matching field vectors. The method obtains the spatial matching between the two images in a global spatial extent and with sub-pixel accuracy. Results of the method on real images and high non-rigid artificial deformation proves the validity of the approach. Its extension to a volumetric approach is also suggested.