Object Representation and Segmentation

Tutorial on Object Representation Much of our work on segmentation is based on medial geometry and the statistics of various object representations       Object representations provide a basis for segmentation by deformable models. We have segmentation approaches by deforming m-reps , deforming spherical harmonics representations, and deforming dense positional grids using diffeomorphic fluid deformations. We can also simulate intra-patient deformations via multigrid finite element models using m-rep based meshing.     We also work on segmentation via geometric flows. A general tool is provided by SNAP. Geometric flows (PowerPoint file) together with statistical pattern recognition is applied for the segmentation of brain tumors     Medial representations are used not only for segmentation by deformable object models but also by direct medial analysis of image data. The resulting loci are called "cores" (PowerPoint file). We have a project on core extraction and have developed a method of locating image objects via their medial features using core atom statistics.     Segmentation, registration, visualization, tracking, characterization, and quantification of tubular networks with an emphasis on networks of blood vessel for disease detection and diagnosis and surgical planning, guidance, and outcome assessment has also been a large area of our work.     We are beginning to be concerned not only of segmentation within single images but also with spatiotemporal segmentation , i.e., segmentation within time series of images.     Segmentation methods must be verified (PowerPoint file). Part of our research focuses on the development of methodology for the verification of segmentation.     Lately we have begun development efforts concerning the processing and visualization of magnetic resonance diffusion tensor imaging (DTI) applied to brain fiber tract analysis.