Image warping through geometric model decomposition

The Edinburgh Mouse Atlas Project (EMAP) at the Medical Research Council (MRC) Human Genetics Unit (Edinburgh) is developing a spatio-temporal framework of mouse development from the single cell stage through to birth. This consists of a time series of 3D (voxel) models of mouse embryos which define the spatio-temporal framework onto which is mapped spatial data such as the recognized anatomy and the patterns of gene expression from many embryos. A multi-modal warping algorithm has been devised to achieve this mapping. Matching sections are found in the atlas (target) and experiment (source) images, from which tie-points are computed using a novel algorithm. Geometric models, in which connectivity is represented explicitly, are built by extracting maximal gradient edges from the source and target images. The source model is then progressively decomposed into smaller connected fragments, while each of the fragments is affine registered to the target model using an iterative closest point (ICP) algorithm. From the target model, the decomposed fragments of the source model and the associated affine transforms, tie-points are computed. These are then used to define a radial basis function warp transformation.