Abstract
Statistical shape models have become an indispensable tool for image analysis. The use of shape models is especially popular in computer vision and medical image analysis, where they were incorporated as a prior into a wide range of different algorithms. In spite of their big success, the study of statistical shape models has not received much attention in recent years. Shape models are often seen as an isolated technique, which merely consists of applying Principal Component Analysis to a set of example data sets. In this thesis we revisit statistical shape models and discuss their construction and applications from the perspective of machine learning and kernel methods. The shapes that belong to an object class are modeled as a Gaussian Process whose parameters are estimated from example data. This formulation puts statistical shape models in a much wider context and makes the powerful inference tools from learning theory applicable to shape modeling. Furthermore, the formulation is continuous and thus helps to avoid discretization issues, which often arise with discrete models. An important step in building statistical shape models is to establish surface correspondence. We discuss an approach which is based on kernel methods. This formulation allows us to integrate the statistical shape model as an additional prior. It thus unifies the methods of registration and shape model fitting. Using Gaussian Process regression we can integrate shape constraints in our model. These constraints can be used to enforce landmark matching in the fitting or correspondence problem. The same technique also leads directly to a new solution for shape reconstruction from partial data. In addition to experiments on synthetic 2D data sets, we show the applicability of our methods on real 3D medical data of the human head. In particular, we build a 3D model of the human skull, and present its applications for the planning of cranio-facial surgeries.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Disclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.